It is a vs code fork. There were some UI glitches. Some usability was better. Cursor has some real annoying usability issues - like their previous/next code change never going away and no way to disable it. Design of this one looks more polished and less muddy.
I was working on a project and just continued with it. It was easy because they import setting from cursor. Feels like the browser wars.
Anyway, I figured it was the only way to use gemini 3 so I got started. A fast model that doesn't look for much context. Could be a preprompt issue. But you have to prod it do stuff - no ambition and a kinda offputting atitude like 2.5.
But hey - a smarter, less context rich Cursor composer model. And that's a complement because the latest composer is a hidden gem. Gemini has potential.
So I start using it for my project and after about 20 mins - oh, no. Out of credits.
What can I do? Is there a buy a plan button? No? Just use a different model?
What's the strategy here? If I am into your IDE and your LLM, how do I actually use it? I can't pay for it and it has 20 minutes of use.
I switched back to cursor. And you know? it had gemini 3 pro. Likely a less hobbled version. Day one. Seems like a mistake in the eyes of the big evil companies but I'll take it.
Real developers want to pay real money for real useful things.
Google needs to not set themselves up for failure with every product release.
If you release a product, let those who actually want to use it have a path to do so.
So I start using it for my project and after about 20 mins - oh, no. Out of credits.
I didn't even get to try a single Gemini 3 prompt. I was out of credits before my first had completed. I guess I've burned through the free tier in some other app but the error message gave me no clues. As far as I can tell there's no link to give Google my money in the app. Maybe they think they have enough.
After switching to gpt-oss:120b it did some things quite well, and the annotation feature in the plan doc is really nice. It has potential but I suspect it's suffering from Google's typical problem that it's only really been tested on Googlers.
EDIT: Now it's stuck in a loop repeating the last thing it output. I've seen that a lot on gpt-oss models but you'd think a Google app would detect that and stop. :D
EDIT: I should know better than to beta test a FAANG app by now. I'm going back to Codex. :D
VSCode is Electron based which, yes, is based on Chromium. But the page you link to isn't about that, its about using VSCode as dev environment for working on Chromium, so I don't know why you linked it in this context.
> Both are based on khtml. We could be living in a very different world if all that effort stayed inside the KDE ecosystem
How so?
Do you think thousands of googlers and apple engineers could be reasonably managed by some KDE opensource contributors? Or do you imagine google and apple would have taken over KDE? (Does anyone want that? Sounds horrible.)
Which came from "the KDE HTML Widget" AKA khtmlw. Wonder if that's the furthest we can go?
> if all that effort stayed inside the KDE ecosystem
Probably nowhere, people rather not do anything that contribute to something that does decisions they disagree with. Forking is beautiful, and I think improves things more than it hurts. Think of all the things we wouldn't have if it wasn't for forking projects :)
On the other hand if that had stopped google from having a browser they push into total dominance with the help of sleazy methods, maybe that would have been better overall.
I still prefer a open source chromium base vs a proprietary IE (or whatever else) Web Engine dominating.
(Fixing IE6 issues was no fun)
Also I do believe, the main reason chrome got dominance is simply because it got better from a technical POV.
I started webdev on FF with firebug. But at some point chrome just got faster with superior dev tools. And their dev tools kept improving while FF stagnated and rather started and maintained u related social campaigns and otherwise engaged with shady tracking as well.
> I still prefer a open source chromium base vs a proprietary IE (or whatever else) Web Engine dominating.
Okay but that's not the tradeoff I was suggesting for consideration. Ideally nothing would have dominated, but if something was going to win I don't think it would have been IE retaking all of firefox's ground. And while I liked Opera at the time, that takeover is even less likely.
> Also I do believe, the main reason chrome got dominance is simply because it got better from a technical POV.
Partly it was technical prowess. But google pushing it on their web pages and paying to put an "install chrome" checkbox into the installers of unrelated programs was a big factor in chrome not just spreading but taking over.
We might not have had Mozilla/Phoenix/Firefox in the first place if so either, who I'd like to think been a net-positive for the web since inception. At least I remember being saved by Firefox when the options were pretty much Internet Explorer or Opera on a Windows machine.
that's a bit misleading. it was based on webcore which apple had forked from khtml. however google found apple's addition to be a drag and i think very little of it (if anything at all, besides the khtml foundation) survived "the great cleanup" and rewrite that became blink. so actually webkit was a just transitional phase that led to a dead end and it is more accurate to say that blink is based on khtml.
Firstly, the barrier to entry lower for people to take web experience and create extensions, furthering the ecosystem moat for Electron-based IDEs.
Even more importantly, though, the more we move towards "I'm supervising a fleet of 50+ concurrent AI agents developing code on separate branches" the more the notion of the IDE starts to look like something you want to be able to launch in an unconfigured cloud-based environment, where I can send a link to my PM who can open exactly what I'm seeing in a web browser to unblock that PR on the unanswered spec question.
Sure, there's a world where everyone in every company uses Zed or similar, all the way up to the C-suite.
But it's far more likely that web technologies become the things that break down bottlenecks to AI-speed innovation, and if that's the case, IDEs built with an eye towards being portable to web environments (including their entire extension ecosystems) become unbeatable.
> Firstly, the barrier to entry lower for people to take web experience and create extensions, furthering the ecosystem moat for Electron-based IDEs.
The last thing I want is to install dozens of JS extensions written by people who crossed that lower barrier. Most of them will probably be vibe coded as well. Browser extensions are not the reason I use specific browsers. In fact, I currently have 4 browser extensions installed, one of which I wrote myself. So the idea that JS extensions will be a net benefit for an IDE is the wrong way of looking at it.
Besides, IDEs don't "win" by having more users. The opposite could be argued, actually. There are plenty of editors and IDEs that don't have as many users as the more popular ones, yet still have an enthusiastic and dedicated community around them.
VS Code is technically an Electron app, but it's not the usual lazy resource hog implementation like Slack or something. A lot of work went into making it fast. I doubt you'll find many non-Electron full IDEs that are faster. Look at Visual Studio, that's using a nice native framework and it runs at the speed of fossilized molasses.
VSCode has even less features than Emacs, OOTB. Complaining about full IDEs slowness is fully irrelevant here. Full IDEs provide an end to end experience in implementing a project. Whatever you need, it's there. I think the only plugins I've installed on Jetbrains's ones is IdeaVim and I've never needed something else for XCode.
It's like complaining about a factory's assembly line, saying it's not as portable as the set of tools in your pelican case.
In 2025, you really picked Emacs as the hill to die on? Who is under 30 who cares about Emacs in 2025? Few. You might as well argue that most developers should be using Perl 6.
> the only plugins I've installed on Jetbrains's ones
By default, JetBrains' IntelliJ-based IDEs have a huge number of plug-ins installed. If you upgrade from Community Edition to a paid license, the number only increases. Your comment is slightly misleading to me.
Just wait until vi steps into the room. Perhaps we can recreate the Usenet emacs vs vi flame wars. Now, if only '90's me could see the tricked out neovim installs we have these days.
Please take a look at the Emacs documentation sometimes.
VSCode is more popular, which makes it easy to find extensions. But you don’t see those in the Emacs world because the equivalent is a few lines of config.
So what you will see are more like meta-extensions. Something that either solve a whole class of problems, could be a full app, or provides a whole interaction model.
VS Code is plenty fast enough. I switched to Zed a few months back, and it's super snappy. Unless you're running on an incredibly resource constrained machine, it mostly comes down to personal preference.
I have always found JetBrains stuff super snappy. I use neovim as a daily driver but for some projects the inference and debugging integration in JetBrains is more robust.
It funny that despite how terrible, convoluted and maladapted web tech is for displaying complex GUIs it still gradually ate lunch of every native component library and they just couldn't innovate to keep up on any front.
Amazon just released OS that uses React Native for all GUI.
It's easy to design bad software and write bad code. Like the old saying: "I didn't have time to write you a short letter, so I wrote you a long one". Businesses don't have time to write good and nice software, so they wrote bad one.
I didn't really mean Electron, but rather unholy amalgam of three languages, each with 20 years of "development", which mostly consisted of doing decrapifying and piling up new (potentially crappy) stuff. Although Electron with UI context and system (backend? background?) context both running js is another can of worms.
Complaining about Electron is an ideological battle, not a practical argument. The people who push these arguments don’t care that it actually runs very well on even below average developer laptops, they think it should have been written in something native.
The word "developer" is doing a lot of work there spec-wise.
The extent to which electron apps run well depends on how many you're running and how much ram you had to spare.
When I complain about electron it has nothing to do with ideology, it's because I do run out of memory, and then I look at my process lists and see these apps using 10x as much as native equivalents.
And the worst part of wasting memory is that it hasn't changed much in price for quite a while. Current model memory has regularly been available for less than $4/GB since 2012, and as of a couple months ago you could get it for $2.50/GB. So even a 50% boost in use wipes out the savings since then. And sure the newer RAM is a lot faster, but that doesn't help me run multiple programs at the same time.
> The word "developer" is doing a lot of work there spec-wise.
Visual Studio Code is a developer tool, so there’s no reason to complain about that.
I run multiple Electron apps at a time even on low spec machines and it’s fine. The amount of hypothetical complaining going on about this topic is getting silly.
You know these apps don’t literally need to have everything resident in RAM all the time, right?
Ease of writing and testing extensions is actually the cause why Electron won IDE wars.
Microsoft made a great decision to jump on the trend and just pour money to lap Atom and such in optimization and polish.
Especially when you compare it to Microsoft effort for desktop. They acumulated several more or less component libraries over they years and I still prefer WinForms.
If you want electron app that doesn't lag terribly, you'll end up rewriting ui layer from scratch anyway. VSCode already renders terminal on GPU and GPU-rendered editor area is in experimental. There will soon be no web ui left at all
> If you want electron app that doesn't lag terribly
My experience with VS Code is that it has no perceptible lag, except maybe 500ms on startup. I don't doubt people experience this, but I think it comes down to which extensions you enable, and many people enable lots of heavy language extensions of questionable quality. I also use Visual Studio for Windows builds on C++ projects, and it is pretty jank by comparison, both in terms of UI design and resource usage.
I just opened up a relatively small project (my blog repo, which has 175 MB of static content) in both editors and here's the cold start memory usage without opening any files:
- Visual Studio Code: 589.4 MB
- Visual Studio 2022: 732.6 MB
update:
I see a lot of love for Jetbrains in this thread, so I also tried the same test in Android Studio: 1.69 GB!
That easily takes the worst designed benchmark in my opinion.
Have you tried Emacs, VIM, Sublime, Notepad++,... Visual Studio and Android Studio are full IDEs, meaning upon launch, they run a whole host of modules and the editor is just a small part of that. IDEs are closer to CAD Software than text editors.
- notepad++: 56.4 MB (went gray-window unresponsive for 10 seconds when opening the explorer)
- notepad.exe: 54.3 MB
- emacs: 15.2 MB
- vim: 5.5MB
I would argue that notepad++ is not really comparable to VSCode, and that VSCode is closer to an IDE, especially given the context of this thread. TUIs are not offering a similar GUI app experience, but vim serves as a nice baseline.
I think that when people dump on electron, they are picturing an alternative implementation like win32 or Qt that offers a similar UI-driven experience. I'm using this benchmark, because its the most common critique I read with respect to electron when these are suggested.
It is obviously possible to beat a browser-wrapper with a native implementation. I'm simply observing that this doesn't actually happen in a typical modern C++ GUI app, where the dependency bloat and memory management is often even worse.
I never understand why developers spend so much time complaining about "bloat" in their IDEs. RAM is so incredibly cheap compared to 5/10/15/20 years ago, that the argument has lost steam for me. Each time I install a JetBrains IDE on a new PC, one of the first settings that I change is to increase the max memory footprint to 8GB of RAM.
> RAM is so incredibly cheap compared to 5/10/15/20 years ago
Compared to 20 years ago that's true. But most of the improvement happened in the first few years of that range. With the recent price spikes RAM actually costs more today than 10 years ago. If we ignore spikes and buy when the cycle of memory prices is low, DDR3 in 2012 was not much more than the price DDR5 was sitting at for the last two years.
Anyone saying that Java-based Jetbrains is worse than Electron-based VS Code, in terms of being more lightweight, is living in an alternate universe which can’t be reached by rational means.
The anti-Electron meme is a vocal minority who don’t realize they’re a vocal minority. It’s over represented on Hacker News but outside of HN and other niches, people do not care what’s under the hood. They only care that it works and it’s free.
I used Visual Studio Code across a number of machines including my extremely underpowered low-spec test laptop. Honestly it’s fine everywhere.
Day to day, I use an Apple Silicon laptop. These are all more than fast enough for a smooth experience in Visual Studio Code.
At this point the only people who think Electron is a problem for Visual Studio Code either don’t actually use it (and therefore don’t know what they’re talking about) or they’re obsessing over things like checking the memory usage of apps and being upset that it could be lower in their imaginary perfect world.
What a strange claim. For enterprise Java, is there is a serious alternative in 2025? And, Rider is slowly eating the lunch of (classic) Visual Studio for C# development. I used it again recently to write an Excel XLL plug-in. I could not believe how far Rider has come in 10 years.
I just checked and I don’t even have the JVM installed on my machine. It seems like Java is dead for consumer applications. Not saying that’s why they aren’t popular but I’m sure it doesn’t help.
At best, that's subjective, but it's fact that JetBrains is comically far behind when it comes to AI tooling.
They have a chance to compete fresh with Fleet, but they are not making progress on even the basic IDE there, let alone getting anywhere near Cursor when it comes to LLM integration.
JetBrains' advantage is that they have full integration and better understanding of your code. WebStorm works better with TypeScript than even Microsoft's own creation. This all translates into AI performance
Have you actually given them a real test yet - either Junie or even the baseline chat?
I believe our definitions of "winning the IDE wars" are very, very different. For one thing, using "user count" as a metric for this like using "number of lines of code added" in a performance review. And even if that was part of the metric, people who use and don't absolutely fall in love with it, so much so that they become the ones advocating for its use, are only worth a tiny fraction of a "user".
neovim won the IDE wars before it even started. Zed has potential. I don't know what IntelliJ is.
It started as a modernized Eclipse competitor (the Java IDE) but they've built a bunch of other IDEs based on it. Idk if it still runs on Java or not, but it had potential last I used it about a decade ago. But running GUI apps on the JVM isn't the best for 1000 reasons, so I hope they've moved off it.
Android Studio is built on the IntelliJ stack. Jetbrains just launched a dedicated Claude button (the button just opens up claude in the IDE, but there are some pretty neat IDE integrations that it supports, like being able to see the text selection, and using the IDE's diff tool). I wonder if that's why Google decided to go VS code?
Uh, isn't that the regular Claude code extension that's been available for ages at this point? Not jetbrains but anthropics own development?
As a person paying for the jetbrains ultimate package (all ides), I think going with vscode is a very solid decision.
The jetbrains ides still have various features which I always miss whenever I need to use another IDE (like way better "import" suggestions as an easy to understand example)... But unless you're writing in specific languages like Java, vscode is way quicker and works just fine - and that applies even more to agentic development, where you're using these features less and less...
Jetbrains IDEs are all based on the JVM - and they work better than VSCode or the full Visual Studio for me. It's the full blown VS (which has many parts written in C++) that is the most sluggish of them all.
Since you last used IntelliJ "about a decade ago", what do you use instead?
> But running GUI apps on the JVM isn't the best for 1000 reasons, so I hope they've moved off it.
What would you recommend instead of Swing on JVM? Since you have "1000 reasons", it should easy to list a few here. As a friendly reminder, they would need to port (probably) millions of lines of Java source code to whatever framework/language you select. The only practical alternative I can think of would be C++ & Qt, but the development speed would be so much slower than Java & Swing.
Also, with the advent of wildly modern JVMs (11+), the JIT process is so insanely good now. Why cannot a GUI be written in Swing and run on the JVM?
Lol the second I saw the antigravity release I thought "there's no way I'm using that, they will kill it within a year". Looks like they're trying to kill it at birth.
Exactly my reaction. Every time I've used something from Google, it ends up dead in a few years. Life is too short to waste so many years learning something that is destined to die shortly
These are just extended press releases, for marketing and management layers, who don't have to use these things themselves, but can look good, when talking about it.
agree but at the same time there's not too much lock in with these IDEs these days and switching is very easy. Especially since they're all VSCode forks
> What's the strategy here? If I am into your IDE and your LLM, how do I actually use it? I can't pay for it and it has 20 minutes of use.
I wonder how much Google shareholders paid for that 20 minutes. And whether it's more or less than the corresponding extremely small stock price boost from this announcement.
speaking of paying for LLMs, am i doing something wrong? i paid cursor $192 for a year of their entry level plan and i never run out of anything. I code professionally, albeit i'm at the stage where it's 80% product dev in finding the right thing to build.
Is there another world where $200/m is needed to run hundreds of agents or something?
When did you pay for it? There was a time when its limits were very generous. If you bought an annual plan at that time then you will continue with that until renewal. Or, alternatively, you’re using the Auto model which is still apparently unlimited. That’s going away.
It’s very easy to run into limits if you choose more expensive models and aren’t grandfathered.
if it were true, it would be a big miss to not point that out when you run out of credit, in their pricing page, or anywhere in their app.
I should also mention that the first time I prompted it, I got a different 'overloaded' type out of credit message. The one I got at the end was different.
I've rotated on paying the $200/month plans with Anthropic, Cursor, and OpenAI. But never Google's. They have maybe the best raw power in their models - smartest, and extremely fast for what they are. But they always drop the ball on usability. Both in terms of software surrounding the model and raw model attitude. These things matter.
With vendor lock-in to Google's AI ecosystem, likely scraping/training on all of your code (regardless of whatever their ToS/EULA says), and being blocked from using the main VS Code extensions library.
Thank you for saying what this entire blog post doesn't. It's actually disrespectful of Google to launch this without even a mention of the fact that it is based on VSCode.
Well, Kate has been around as an KDE based advanced text editor for nearly 2 decades now - its base feature set isn't too different from a base VS Code installation. And there's also KDevelop as a more full featured IDE.
It's a good point, and in fact I went and looked at the original announcement of VS Code and it appears that Microsoft didn't credit Chromium or Electron back then either. I guess big companies are allergic to crediting other big companies.
> given how many other companies have "created browsers" that are just Chromium forks and rubbed Google the wrong way
Has there been any indication that these folks are "rubbing Google the wrong way"? I think Chromium, as a project, is actually very happy that more people are using their engine.
It also feels like they couldn't use the GOOGLE ANTIGRAVITY logo enough times in this blog post. Gigantic image with the logo and a subtitle, plastered over and over again.
I no longer bother reading their press releases. I'd much rather read the comments and threads like these to get the real story. And I say that as a former googler.
It's so obvious from even just the vague screenshots that are hidden somewhere on the site that it's a VSCode fork, that I suppose I can see why they've tried to obfuscate that as much as possible.
VSCode isn't a Chromium fork, it's an Electron app. Utilizing something is different than making a derivative of it. For example, an empty "Hello World" Electron app wouldn't have any value for an app developer, but creating a web browser derived from Chromium means you've already finished 99.9% of the work.
Why credit? Come on, the world has moved on from 1990s-era 4-clause BSD licenses. If you recall, the 4-clause BSD license states that all advertising materials must display an acknowledgement. It’s widely considered to be a mistake and nobody uses this license any more. Not because of legal reasons (incompatibility with GPL) but because it is madness to require so many acknowledgements. Stallman was right.
Yes, madness. VS Code wasn’t developed entirely by Microsoft. It uses plenty of other open source libraries. Why is it that VS Code should be acknowledged but not the underlying V8 engine, or Chromium, or WebKit or KHTML?
Stallman said that in 1997 there were 75 acknowledgements in a single piece of software. With today’s trend of micro libraries on npm, there will be at least thousands of acknowledgements in one piece of software.
Google is going to win AI and kill all the other market participants.
They have the revenues to support all of this.
They spent time learning from all the players and can now fast follow into every market. Now they're fast and nimble and are willing to clone other products wholesale, fork VSCode, etc.
They're developing all of this, meanwhile Pichai is calling it a "bubble" to put a chill on funding (read: competition). It's not like Google is slowing down.
We had a chance to break them up with regulation, and we didn't. Now they're going to kill every market participant.
This isn't healthy. We have an invasive species in the ecology eating up all the diverse, healthy species.
a16z and YC must hate this. It puts a cap on their returns.
As engineers, you should certainly hate this. Google does everything it can to push wages down. Layoffs, offshoring, colluding with competitors. Fewer startups mean fewer rewards for innovation capital and more accrual to the conglomerate taxing the entire internet.
Chrome, Android, Search, Ads, YouTube, Cloud, Workspace, Other Bets, and AI/Deepmind need to be split into separate companies.
Google? Push wages down? Google is mostly known for paying top of market to keep a zoo of engineers whose only output is blog posts about how smart they are because they solved a problem they also caused.
(presumably because if they touch the ad system it might break)
> a16z and YC must hate this. It puts a cap on their returns.
And a16z's main business is investing in financial scams.
You mean like web search, webmail, internet ads, maps, calendars, browsers, smartphone operating systems, online document editing, and translation? I mean, I'm not even including stuff they acquired early like YouTube. Google was the most feared company for a decade or more for a good reason: they absolutely devoured competition in what were thought to be mature markets.
Putting aside that several of these were acquisitions, these are all great examples of things where Google introduced something for free because it would make the money through advertising, both directly and through ecosystem effects. Even the paid enterprise versions of these services were a tiny % of Google's overall gross revenue.
Prior to the push into Cloud computing, Ad revenue was well over 90% of all Google gross income, and Cloud was the first big way they diversified. GCP is definitely a credible competitor these days, but it did not devour AWS. Other commercial Google services didn't even become credible competitors, e.g. Google Stadia was a technically exceptional platform that got nowhere with customers.
The question now is whether Google carves out an edge in AI that makes it profitable overall, directly or strategically. Like many companies, there seems to be a presumption of potentially infinite upside, which is what it would take to justify the astronomical costs.
Google’s main ability is to win by pure technical prowess. They hire a lot of bright engineers. Google Search won over Altavista by pure algorithms. Google Docs (and Writely) were way more feature complete than competitors.
You love a Google product because of its features but never actually because of the product itself. But you can’t win everything by engineering and sometimes Google struggles with the product side.
I'm not sure you can call Docs (Writely) and Android acquisitions though. Android was an OS for cameras and Writely was an experimental rich text editor, not a word processor.
It's not like Youtube where they legitimately bought their way to dominance. And I'd argue that even in the case of DoubleClick, google was already dominating the search advertising market when they bought DoubleClick to consolidate their dominance.
> Plaintiffs maintain that Google has monopoly power in the product market for general
search services in the United States.
> According to Plaintiffs, Google has a dominant and durable
share in that market (general search), and that share is protected by high barriers to entry.
> Google counters that there is no such thing as a product market for general search services.
> What exists instead, Google insists, is a broader market for query response.
(+ yes obviously, products like Sheets or Maps were amazing, and are still very much the best.
It was a joke to say that even Google denies its own success, the same way as the earlier comment).
Just the Eclipse legacy and development approach perhaps. Eclipse is 24 years old and its codebase stems from VisualAge before that, so I imagine it's ... idiosyncratic. Probably the most successful thing built on OSGi, though I dont imagine Theia brought any of that along.
At least for a while, Eclipse seemed like an Architect Astronaut's happy fever dream: actual bits if implementations hidden behind 5 to 10 nested interfaces and facades wherever I looked. I remember that at one point that I wanted to debug some weird behavior in the Plugin Development Kit and after sinking about half a day into exploring the source code, I didn't even see a single line of code that was actually doing anything meaningful. It was quite shocking to my old junior level self.
I am not fan of Eclipse, mostly due to bad experience with it, but this is excellent idea, if more people and companies would invest in it's development, we would have alternative to VS Code.
Interesting that they include non-Gemini models. Both Claude and GPT oss are both on Google Cloud, so I assume that Antigravity is using GC as the provider and not making API calls to Anthropic or OpenAI.
As somebody who worked on two IDEs which didn't fork VSCode but still used Monaco for code editing views, I think forking VSCode is almost always the right solutions for a new IDE. You get extensions, familiarity and most importantly, don't waste valuable time on the boring stuff which VSCode has already implemented.
Nothing bad with using code other people made open. Our whole industry is built on this.
forking vscode? simple. extensions not so simple. they are controlled by microsoft. without them you’ll run into continual papercuts as a vendor who has forked vscode.
Because if they're just an extension they're stuck with whatever rules Microsoft makes up, and Google is no stranger to using this leverage against others.
Because there are plenty of good reasons why you may want to modify/extend the code and the look and feel beyond what an extension would let you do.
I never understood why people scoff at VS Code forks. I'd honestly tend to be more skeptical of new editors that don't fork VS Code, because then they're probably missing a ton of useful capabilities and are incompatible with all the VSC extensions everyone's gotten used to.
Native app dev is covered in red tape and puts you at the mercy of Apple etc. It's unfortunate that things are so inefficient now, but competition is good, and native platforms can get good.
> Native app dev is covered in red tape and puts you at the mercy of Apple etc.
... Aren't we talking about a programming IDE here? When did mobile become anything like the primary market for that? Are people expected to sit around for hours inputting symbols with an OSK?
I tried writing a native Windows app using WinUI 3.
I wasted a day on trying to get some PNGs to render correctly, but no matter the config I used, the colors came out wrongly oversaturated.
I used Tauri with a WebView, and the app was rendering the images perfectly fine. On top of that the UI looked much better, and I was done in half the time I spend trying to fix the rendering issue in WinUI 3.
Making a VSCode fork is probably the wrong direction at this point in time. The future of agentic coding should need less support for code editor related functionality, and could eventually primarily support viewing code rather than editing code. There's a lot more flexibility in UI starting from scratch, and personally I want to see a UI that allows flexible manipulation of context and code changes with multiple agents.
GitHub is building a UI like this. I like it. I sometimes need the full IDE, but plenty of times don't. It's nice to be able to easily see what the agent is up to and converse with it in real-time while reviewing it's outputs.
Whenever I have a model fix something new I ask it to update the markdown implementation guides I have in the docs folder in my projects. I add these files to context as needed. I have one for implementing routes and one for implementing backend tests and so on.
They then know how to do stuff in the future in my projects.
They still aren't learning. You're learning and then telling them to incorporate your learnings. They aren't able to remember this so you need to remind them each day.
That sounds a lot like '50 First Dates' but for programming.
Yes, this is something people using LLMs for coding probably pick up on the first day. They're not "learning" as humans do obviously. Instead, the process is that you figure out what was missing from the first message you sent where they got something wrong, change it, and then restart from beginning. The "learning" is you keeping track of what you need to include in the context, how that process exactly works, is up to you. For some it's very automatic, and you don't add/remove things yourself, for others is keeping a text file around they copy-paste into a chat UI.
This is what people mean when they say "you can kind of do "learning" (not literally) for LLMs"
Yeah, if my screwdriver undid the changes I just made to my mower, constantly ignored my desire to unscrew screws and instead punched a hole in my carb - I'd be throwing that screwdriver in the garbage.
- don't have career growth that you can feel good about having contributed to
Humans are on the verge of building machines that are smarter than we are. I feel pretty goddamned awesome about that. It's what we're supposed to be doing.
- don't have a genuine interest in accomplishment or team goals
Easy to train for, if it turns out to be necessary. I'd always assumed that a competitive drive would be necessary in order to achieve or at least simulate human-level intelligence, but things don't seem to be playing out that way.
- have no past and no future. When you change companies, they won't recognize you in the hall.
Or on the picket line.
- no ownership over results. If they make a mistake, they won't suffer.
Good deal. Less human suffering is usually worth striving for.
> Humans are on the verge of building machines that are smarter than we are. I feel pretty goddamned awesome about that. It's what we're supposed to be doing.
It's also the premise of The Matrix. I feel pretty goddamned uneasy about that.
(Shrug) There are other sources of inspiration besides dystopic sci-fi movies. There's the Biblical story of the Tower of Babel, for instance. Better not work on language translation, which after all is how the whole LLM thing got started.
Sometimes fiction went in the wrong direction. Sometimes it didn't go far enough.
In any case, the matrix wasn't my inspiration here, but it is a pithy way to describe the concept. It's hard to imagine how humans maintain relevancy if we really do manage to invent something smarter than us. It could be that my imagination is limited though. I've been accused of that before.
> Humans are on the verge of building machines that are smarter than we are.
You're not describing a system that exists. You're describing a system that might exist in some sci-fi fantasy future. You might as well be saying "there's no point learning to code because soon the rapture will come".
That particular future exists now, it's just not evenly distributed. Gemini 2.5 Pro Thinking is already as good at programming as I am. Architecture, probably not, but give it time. It's far better at math than I am, and at least as good at writing.
They can usually write code, but not that well. They have lots of energy and little to say about architecture and style. Don't have a well defined body of knowledge and have no experience. Individual juniors don't change, but the cast members of your junior cohort regularly do.
I actually do find there is a subset of meetings that are far more productive on Zoom. We can be voice chatting on one screen, share another screen and both be able to type, record notes, pull up side research without interrupting the conversation. It's a bit closer to co-working than a meeting but it hits a sweetspot for me.
I used to be really excited about "agents" when I thought people were trying to build actual agents like we've been working on in the CS field for decades now.
It's clear now that "agents" in the context of "AI" is really about answering the question "How can we make users make 10x more calls to our models in a way that makes it feel like we're not just squeezing money out of them?" I've seen so many people that think setting some "agents" of on a minutes to hours long task of basically just driving up internal KPIs at LLM providers is cutting edge work.
The problem is, I haven't seen any evidence at all that spending 10x the number of API calls on an agent results in anything closer to useful than last year when people where purely vibe coding all the time. At least then people would interactively learn about the slop they were building.
It's astounding to watch a coworker walk though through a PR with hundreds of added new files and repeatedly mention "I'm not sure if these actually work, but it does look like there's something here".
Now I'm sure I'll get some fantastic "no true Scotsman" replies about how my coworkers must not be skilled enough or how they need to follow xyz pattern, but the entire point of AI was to remove the need for specialize skills and make everyone 10x more productive.
Not to mention that the shift in focus on "agents" is also useful in detracting from clearly diminishing returns on foundation models. I just hope there are enough people that still remember how to code (and think in some cases) to rebuild when this house of cards falls apart.
> but the entire point of AI was to remove the need for specialize skills and make everyone 10x more productive.
At least for programming tools, for everything (well, the vast majority, at least) that is sold that way—since long before generative AI—it actually succeeds or fails based not on whether it eliminates need for specialized skills and makes everyone more productive, but whether it further rewards specialized skills, and makes the people who devote time to learning it more productive than if they devoted the same time to learning something else.
I am essentially in this exact role. The junior developers simply don't have the experience to evaluate the output of the agents. You wind up with a lot of slop in PRs. People can't justify why they did something. I've seen whole PRs closed, work redone, and opened anew because they were 70% garbage. Every other comment was asking "why is this here? it has nothing to do with the ticket."
Sadly, this is not sustainable and I am not sure what I'm going to do.
I enjoy getting into a good flow state and pounding out clever and elegant code but watching a good LLM generate code according to my specs and refining it is also enjoyable. I've been burning through $250 of free Claude Code Web credits and having multiple workers running at the same time is fun.
They have a capacity to "learn", it's just WAY MORE INVOLVED than how humans learn.
With a human, you give them feedback or advice and generally by the 2nd or 3rd time the same kind of thing happens they can figure it out and improve. With an LLM, you have to specifically setup a convoluted (and potentially financially and electrical power expensive) system in order to provide MANY MORE examples of how to improve via fine tuning or other training actions.
Depending on your definition of "learn", you can also use something akin to ChatGPT's Memory feature. When you teach it something, just have it take notes on how to do that thing and include its notes in the system prompt for next time. Much cheaper than fine-tuning. But still obviously far less efficient and effective than human learning.
I think it’s reasonable to say that different approaches to learning is some kind of spectrum, but that contemporary fine tuning isn’t on that spectrum at all.
> With an LLM, you have to specifically setup a convoluted (and potentially financially and electrical power expensive) system in order to provide MANY MORE examples of how to improve via fine tuning or other training actions.
The only way that an AI model can "learn" is during model creation, which is then fixed. Any "instructions" or other data or "correcting" you give the model is just part of the context window.
As a team lead, working with people is so... cumbersome. They need time to recharge, lots of encouragement, and a nice place to work in. Give me a coding agent any time!
Well it's a helluva lot faster to make for one. For two, just about everyone knows how to navigate in vscode by now. Reducing the barrier of entry has obvious advantages.
I just opened the app to see what else I can bring up, and while clicking through UI I noticed I had some crappy key bindings extension installed, which apparently caused many of my annoyances.
I've probably installed it very long ago, or even by accident.
For example, I was always annoyed that open file/directory shortcut (one of most common operations) is not assigned and requires mouse interaction -- fixed by disabling the extension.
Go to file shortcuts does something completely different -- fixed by disabling the extensions.
I likely won't adopt Cursor as my main IDE/Editor, but it's miles better than I thought just an hour ago.
Not the person you asked, but I hate how it screws up keyboard shortcuts.
It overrode the delete line shortcut with its own inline chat one, for example.
Decided to ditch it for claude code right after that, since I cannot be bothered to go over the entire list of keyboard shortcuts and see what else it overrode/broke.
I've found that annoying too, but you can always rebind them as you wish. It's only a few new keybinds that get in the way of my muscle memory.
That said I also have moved to CLI agents like Claude Code and Codex because I just find them more convenient and, for whatever reason, more intelligent and more likely to correctly do what I request.
These new editors are trying to differentiate themselves via their AI features. Working on the core editor may waste resources that could have been better spent improving the AI features.
Until someone finally figures out that we need to rethink editors from the ground up to support different sort of operations and editing experience, to better facilitate LLMs doing work as agents.
But we're probably 1-2 years away from there still, so we'll live with skinned-forks, VSCode extensions and TUIs for now.
Zed team is writing their own in-house GUI stack [1] that leverages the computer's GPU with minimal middleware in-between. It's a lot of work short-term but IMO the payoff would be huge if they establish themselves. I imagine they could poke into the user-facing OS sector if their human-agent interaction is smooth. (I have not tried it yet though)
I am very sensitive to input latency and performance but after comparing Zed and VS Code for a while I really couldn't find any reason to stick with Zed. It's been a year or so since I last tried it but VSC just lets me do way more while still, IMO, having a nice, clean UI. I never notice any performance or key input latency with VSC.
> I wonder why they are not trying to fixup something extremely complex that only a handful players managed to get right using gui stacks made with only mobile in mind that are desperately trying to catch up to desktop now
We need a VS Code fork that just exposes more interfaces, and does nothing else. Then all these forks could just use that with power extensions, and it'd force Microsoft to change its behavior.
The issue with Eclipse and that approach is the complexity of mixing plugins to do everything, which kills the UX.
When VSCode started, the differentiator from Atom and Eclipse was that the extension points were intentionally limited to optimize the user experience. But with the introduction of Copilot that wasn’t enough, hence the amount of forks.
I think that the Zed approach of having a common protocol to talk with agents (like a LSP but for agents) is much better. The only thing that holds me from switching to Zed is that so far my experience using it hasn’t been that good (it still has many rough edges)
Microsoft has very specific constraints on what extensions can and can't do, it's not a free for all. They're actively defending their mote by allowing Copilot to do things in a way that extensions couldn't. That's why all the serious contenders make a fork, it's simply not possible to have the same integration otherwise.
Still it would be a lot wiser if all the forkers would do one 'AI-enabled' fork together that exposes all the extras that copilot gets. The barrier for testing would be much lower and all the extension makers would also jump onto the train. Likely MS would finally give in and make all the extras available for everyone. But all the fragmentation only helps MS.
I've had a Github Copilot subscription from work for 1yr+ and switch between the official Copilot and Roo/Kilo Code from time to time. The official Copilot extension has improved a lot in the last 3-6 months but I can't recall ever seeing Copilot do something that Roo/Kilo can't do, or am I missing something obvious?
They're all going to have quite divergent opinions on how to structure the fork and various other design decisions that would inevitably lead to forks of the fork again.
I think forking VS Code is probably the most sensible strategy and I think that will remain the case for many years. Really, I don't think it's changing until AI agents get so ridiculously good that you can vibe code an entire full-featured polished editor in one or a few sessions with an LLM. Then we'll be seeing lots of de novo editors.
The Claude Code extension on VS Code does very little (too little in my opinion). The integration level with agentic functionality provided by Antigravity goes much deeper in my 20 minutes or so of playing with it. The biggest value pieces I see is: Agent Manager window which provides a unified view of all my agents running across all my workspaces (!) where I can quickly approve or respond to followup questions and quickly brings me to the code in context for each agent, additionally, I can select a piece of code and comment on it inline and that comment gets sent to the correct, active agent. These two things alone are items which I have been looking for... Too bad I only have approval to use Claude Code at work. This looks promising.
I don't even know what the Claude Code extension does in vscode. I have it installed but hell if I know what it's doing. I run Claude in one of vscode's terminals, and do everything through there. I do see (sometimes) diffs pop up in the IDE, I guess that's the extent of this integration.
> AIUI the forks are required because Microsoft is gatekeeping functionality used by Copilot from extensions so they can't be used by these agents.
reply
I always wonder how this works legally. VSCode needs to comply with the LGPL (it's based on Chromium/Blink which is LGPL) ; they should provide the entire sources that allow us to rebuild our own "official" VSCode binary
Could you give an example of what they're gatekeeping for Copilot exclusively? I'm kinda confused because Copilot in VS Code isn't exactly a powerhouse of unique features in my experience, it still feels well behind Roo/Kilo Code in most ways I can think of, although much closer to the competition than it was a year ago.
I was going to ask why all these companies choose to fork the entire IDE rather than just writing an extension like every other sane developer, and this response is the most believable reason why.
But Microsoft made VSCode lol, I think being able to gatekeep things like that shouldn’t allow a billion dollar company just reuse all of your code instead of making their own IDE
> 2024: every day a new Chrome fork browser is announced
I think this was more accurate around 2012. My local tech magazine had their own fork and they attached CD with the magazine which included the browser.
This whole blog post is seemingly about Google, not about the user. "Why We Built Antigravity" etc. "We want Antigravity to be the home base for software development in the era of agents" - cool, why would I as the user care about that?
This kind of cynicism is wild to me. Of course most AI products (and products in general) are for end users. Especially for a company like Google--they need to do everything they can to win the AI wars, and that means winning adoption for their AI models.
I do not believe that Google Antigravity is aimed at wooing investors. I believe it is intended to be a genuine superior alternative to Cursor and Kiro etc. and is attempting to provide the best AI coding experience for the average developer.
Most of the other people (so far) in this sub-thread do not think this. They essentially have a conspiratorial view on it.
I agree what you’ve listed makes sense as a product portfolio.
But AI Studio is getting vibe coding tools. AI Studio also has a API that competes with Vertex. They have IDE plugins for existing IDEs to expose Chat, Agents, etc. They also have Gemini CLI for when those don’t work. There is also Firebase Studio, a browser based IDE for vibe coding. Jules, a browser based code agent orchestration tool. Opal, a node-based tool to build AI… things? Stich, a tool to build UIs. Colab with AI for a different type of coding. Notebook LM for AI research (many features now available in Gemini App). AI Overviews and AI mode in search, which now feature a generic chat interface.
Thats just new stuff, and not including all the existing products (Gmail, Home) that have Gemini added.
This is the benefit of a big company vs startups. They can build out a product for every type of user and every user journey, at once.
In "real world" you don't use OpenAI or Anthropic API directly—you are forced to use AWS, GCP, or Azure. Each of these has its own service for running LLMs, which is conceptually the same as using OpenAI or Anthropic API directly, but with much worse DX. For AWS it's called Bedrock, for GCP—Vertex, and for Azure it's AI Foundry I believe. They also may offer complementary features like prompt management, evals, etc, but from what I've seen so far it's all crap.
Jules is the first and only one to add a full API, which I've found very beneficial. It lets you integrate agentic coding features into web apps quite nicely. (In theory you could always hack your own thing together with Claude Code or Codex to achieve a similar effect but a cloud agent with an API saves a lot of effort.)
Remember took my a while early in my career from changing my resume away from saying "I want to do this at my next job and make a lot of money" and towards "here is how I can make money and save costs for your company".
Google didn't learn that lesson here. They are describing why us using Antigravity is good for Google, not why us using Antigravity is good for us.
More accurately, it should be neither about Google nor about the user, but about the product. Describe what the product is and does, don’t make assumptions about the user, and let the user be the judge of it.
On the pricing page it says that for public preview they are offering a free individual plan with "generous rate limits". I gave it an HTML file and asked it to create Jinja templates from it and 2 minutes later (still planning, no additional prompt) I got this:
> Model quota limit exceeded. You have reached the quota limit for this model.
I'm saying it's egregious to expect all users to know the fact that an HTML document, for some reason, uses an enormous amount of context in an LLM designed specifically for working with code.
It's the same problem with OpenRouter's free tiers for a long time. If something is truly $0 and widely available, people will absolutely bleed it dry.
Same here. I tried to build a super simple iOS App in antigravity and I was out of quota before it finished. The whole thing was a couple of files and a few hundred lines of code.
One thing I’ve noticed though that actually coding (without the use of AI; maybe a bit of tab auto-complete) is that I’m actually way faster when working in my domain than I am when using AI tools.
Everytime I use AI tools in my domain-expertise area, I find it ends up slowing me down. Introducing subtle bugs, me having to provide insane amount of context and details (at which point it becomes way faster to do it myself)
Just code and chill man - having spent the last 6 months really trying everything (all these context engineering strategies, agents, CLAUDE.md files on every directory, et, etc). It really easy still more productive to just code yourself if you know what you’re doing.
The thing I love most though - is having discussions with an LLM about an implementation, having it write some quick unit tests and performance tests for certain base cases, having it write a quick shell script, etc. things like this, it’s Amazing and makes me really enjoy programming since I save time and can focus on doing the actual fun stuff
When I'm doing the coding myself, I'm at least making steady progress and the process is predictable. With LLMs, it's a crapshoot. I have to get the AI to understand what I want and may have try again multiple times, many times never succeeding, and end up writing a lot of text anyway. And in between, I'll have to read a lot of code that probably ends up being thrown away or heavily modified.
This probably depends a lot on what kind of project one is working on, though.
But it's like you said, I like using LLMs for completing smaller parts or asking specific kind of help or having conversations about solutions, but for anything larger, it just feels like banging my head to a wall.
One of the core principles of my workflow (inspired by REPL development and some unix tools) is to start with a single file (for a function or the whole project). The I refactor the code to have a better organization and to improve reliability, especially as I'm handling more scenarios (and failure modes).
LLMs are not useful in this workflow, because they are too verbose. Their answers are generic and handle scenarios you don't even support yet. What's useful is good documentation (as in truthful) and the code if it's open.
This approach has worked really well in my career. It gives me KISS and YAGNI for free. And every line of code is purposeful and have a reason to be there.
Devs are starting to realize that the sweet spot for AI support in coding is on a small scale, i.e. extended code completion. Generating huge chunks of code is often not reliable enough except for some niches (such as simple greenfield projects which profit from millions of examples online).
I feel this. Tbh I was excited about most of the previous tech fads I have come across. Sure, people got overzealous with them and also failed to realize that what F*NG needs is not what your three person startup needs, but there were always some good and interesting ideas there.
This just feels... a little too dystopian. Companies hoovered up the entirety more or less of all of our collective thoughts and writings and output and now want to sell it back to us- and I fear that cost is going to be extremely steep.
It's impressive, but at the same time, just feels like its going to somehow be a net detractor to society, and yet I feel I need to keep up with each new iteration or potentially get washed over and left behind by the wave.
I am somewhat fortunate to be towards the top of the pyramid and also in a position where I could theoretically ride off into the sunset, but I fear the societal implications and the pain that is going to come for vast numbers of people.
> The agentic spam is exhausting. I just wanted to code.
Ditto. And we still can.
I've yet to use an "agent", and still use a chat UI to an LLM in Emacs. I rely on these tools for design discussion, rough prototyping, and quick reference, but they still waste my time roughly a quarter of the time I use them. They have gotten better in the last year, though, and I've been able to broaden my reach into stacks and codebases I wouldn't have felt comfortable with before, which is good.
I just have no interest in "agents". I don't want to give these companies more access to my system and data, and I want to review every thing these tools generate. If this makes me slower than a vibe coder, that's intentional. Thankfully, there are still sane people and companies willing to pay me for this type of work, so I'm not worried about being displaced any time soon. Once that happens, I'll probably close up shop, figure out an alternative income stream, and continue coding as a hobby.
> Spin up agents to tackle routine tasks that take you out of your flow, such as codebase research, bug fixes, and backlog tasks.
The software of the future, where nobody on staff knows how anything is built, no one understands why anything breaks, and cruft multiplies exponentially.
After a bunch of people leave the company it's already like nobody knows how anything is built. This seems like a good thing to accelerate understanding a codebase.
it's funny - nervous funny, not haha funny - that you think drawing a real issue like this out into the open would focus an organization on solving it.
You can ask agents to identify and remove cruft. You can ask an agent why something is breaking -- to hypothesize potential causes and test them for validity. If you don't understand how something is built, you can ask the agent to give you an overview of the architecture and then dive into whatever part you want to explore more.
And it's not like any of your criticisms don't apply to human teams. They also let cruft develop, are confused by breakages, and don't understand the code because everyone on the original team has since left for another company.
> you can ask the agent to give you an overview of the architecture and then dive into whatever part you want to explore more.
This is actually a cool use that's being explored more and more. I first saw it in the wiki thing from the devin people, and now google released one as well.
Humans are just better at communicating about their process. They will spend hours talking over architectural decisions, implementation issues, writing technical details in commit messages and issue notes, and in this way they not only debug their decisions but socialize knowledge of both the code and the reasons it came to be that way. Communication and collaboration are the real adaptive skills of our species. To the extent AI can aid in those, it will be useful. To the extent it goes off and does everything in a silo, it will ultimately be ignored - much like many developers who attempt this.
I do think the primary strengths of genai are more in comprehension and troubleshooting than generating code - so far. These activities play into the collaboration and communication narrative. I would not trust an AI to clean up cruft or refactor a codebase unsupervised. Even if it did an excellent job, who would really know?
> Humans are just better at communicating about their process.
I wish that were true.
In my experience, most of the time they're not doing the things you talk about -- major architectural decisions don't get documented anywhere, commit messages give no "why", and the people who the knowledge got socialized to in unrecorded conversations then left the company.
If anything, LLM's seem to be far more consistent in documenting the rationales for design decisions, leaving clear comments in code and commit messages, etc. if you ask them to.
Unfortunately, humans generally are not better at communicating about their process, in my experience. Most engineers I know enjoy writing code, and hate documenting what they're doing. Git and issue-tracking have helped somewhat, but it's still very often about the "what" and not the "why this way".
"major architectural decisions don't get documented anywhere"
"commit messages give no "why""
This is so far outside of common industry practices that I don't think your sentiment generalizes. Or perhaps your expectation of what should go in a single commit message is different from the rest of us...
LLMs, especially those with reasoning chains, are notoriously bad at explaining their thought process. This isn't vibes, it is empiricism: https://arxiv.org/abs/2305.04388
If you are genuinely working somewhere where the people around you are worse than LLMs at explaining and documenting their thought process, I would looking elsewhere. Can't imagine that is good for one's own development (or sanity).
I've worked everywhere from small startups to megacorps. The megacorps certainly do better with things like initial design documents that startups often skip entirely, but even then they're often largely out-of-date because nobody updates them. I can guarantee you that I am talking about common industry practices in consumer-facing apps.
I'm not really interested in what some academic paper has to say -- I use LLM's daily and see first-hand the quality of the documentation and explanations they produce.
I don't think there's any question that, as a general rule, LLM's do a much better job documenting what they're doing, and making it easy for people to read their code, with copious comments explaining what the code is doing and why. Engineers, on the other hand, have lots of competing priorities -- even when they want to document more, the thing needs to be shipped yesterday.
Alright, I'm glad to hear you've had a successful and rich professional career. We definitely agree that engineers generally fail to document when they have competing priorities, and that LLMs can be of use to help offload some of that work successfully.
Your initial comment made it sound like you were commenting on a genuine apples-for-apples comparisons between humans and LLMs, in a controlled setting. That's the place for empiricism, and I think dismissing studies examining such situations is a mistake.
A good warning flag for why that is a mistake is the recent article that showed engineers estimated LLMs sped them up by like 24%, but when measured they were actually slower by 17%. One should always examine whether or not the specifics of the study really applies to them--there is no "end all be all" in empiricism--but when in doubt the scientific method is our primary tool for determining what is actually going on.
But we can just vibe it lol. Fwiw, the parent comment's claims line up more with my experience than yours. Leave an agent running for "hours" (as specified in the comment) coming up with architectural choices, ask it to document all of it, and then come back and see it is a massive mess. I have yet to have a colleague do that, without reaching out and saying "help I'm out of my depth".
Sure, you can ask the agents to "identify and remove cruft" but I never have any confidence that they actually do that reliably. Sometimes it works. Mostly they just burn tokens, in my experience.
> And it's not like any of your criticisms don't apply to human teams.
Every time the limitations of AI are discussed, we see this unfair standard applied: ideal AI output is compared to the worst human output. We get it, people suck, and sometimes the AI is better.
At least the ways that humans screw up are predictable to me. And I rarely find myself in a gaslighting session with my coworkers where I repeatedly have to tell them that they're doing it wrong, only to be met with "oh my, you're so right!" and watch them re-write the same flawed code over and over again.
its so funny watching crusty devs malding over AI. do you all write assembly code for your apps too? because who can trust those magic compiler things right?
AI is just another abstraction that allows you to do 10x with 1x the effort. either you'll learn to master it or you'll become obsolete
How is a compiler and an LLM equivalent abstractions? I'm also seriously doubtful of the 10x claim any time someone brings it up when AI is being discussed. I'm sure they can be 10x for some problems but they can also be -10x. They're not as consistently predictable (and good) like compilers are.
The "learn to master it or become obsolete" sentiment also doesn't make a lot of sense to me. Isn't the whole point of AI as a technology that people shouldn't need to spend years mastering a craft to do something well? It's literally trying to automate intelligence.
This argument warrants introspection for "crusty devs", but also has holes. A compiler is tightly engineered and dependable. I have never had to write assembly because I know that my compiled code 100% represents my abstract code and any functional problems are in my abstract code. That is not true in AI coding. Additionally, AI coding is not just an abstraction over code, but an abstraction over understanding. When my code compiles, I don't need to worry that the compiler misunderstood my intention.
I'm not saying AI is not a useful abstraction, but I am saying that it is not a trustworthy one.
The increasing levels of abstraction work only as long as the abstractions are deterministic (with some limited exceptions - i.e. branch prediction/preloading at CPU level, etc). You can still get into issues with leaky abstractions, but generally they are quite rare in established high->low level language transformations.
This is more akin to manager-level view of the code (who need developers to go and look at the "deterministic" instructions); the abstraction is a lot lot more leaky than high->low level languages.
In the 00s I saw so many C codebases with hand-rolled linked lists where dynamically resized arrays would be more appropriate, "should be big enough" static allocations with no real idea of how to determine that size, etc. Hardly anyone seemed to have a practical understanding of hashes. When you use a higher level language, you get steered towards the practical, fundamental data structures more or less automatically.
even JS doesn't churn as fast as the models powering vibe coding, and that cut & paste node app is still deterministic, compared to what happens when the next version of the model looks at AI-generated code from two years ago...
> Neither engenders user trust in the work that the agent undertook. Antigravity provides context on agentic work at a more natural task-level abstraction, with the necessary and sufficient set of artifacts and verification results, for the user to gain that trust.
I'm going to need an AI summary of this page to even start comprehending this... It doesn't help that the scrolling makes me nauseous, just like real anti-gravity probably would.
"A more intuitive task-based approach to monitoring agent activity, presenting you with essential artifacts and verification results to build trust."
The whole thing around "trust" is really weird. Why would I as a user care about that? It's not as if LLMs are perfect oracles and the only thing between us and a brave new world is blind trust in whatever the LLM outputs.
I'd say that because right now, verifying the LLM output (meaning: not trusting it) is a huge bottleneck (rightfully so). I guess they are trying to convince people that this bottleneck is no more, with this IDE.
Translation: "don't trouble your little brain actually trying to read the code that our model produced. we can show you pictures of the output instead."
It really seems like it's just standardizing into a first-class UI what a lot of people have already been doing.
I don't think I'm the target for this - I already use Claude Code with jj workspaces and a mostly design-doc first workflow, and I don't see why I would switch to this, but I think this could be quite useful for people who don't want to dive in so deep and combine raw tooling themselves.
Sure. Honestly I think you can get the same with git work trees, though I haven't tried.
After a couple iterations on this, I've ended up having claude code vibe-code a helper CLI in Go for me which I can invoke with `ontheside <new-workspace-name> <base-change>` and will
- create a new jj workspace based on the given change
- create a docker container configured with everything my unit tests need to run, my claude code config mounted, and jj configured
- it also sets up a claude code hook to run `jj` (no arguments) every time it changes a file, so that jj does a snapshot
- finishes by starting an interactive claude code session with `--dangerously-skip-permissions`
- it also cleans it all up once I exit the claude code session and fish shell that's running it
With this I can have Claude Code working asynchronously on something, while I can review (or peek) the changes in batch from my main editor by running `jj show <change-id>` / `jj diff -r "..."` (which in my case opens it up in the Goland multi-file diff viewer). I can also easily switch to the change it's working on in my main editor, to make some manual modifications if necessary.
This is, in general, primarily for "in the background async stuff" I want to have it work on. Most of the time I just have a dead-normal claude code session running in my main workspace.
Minor self-plug - if you want, I posted a jj intro article a while ago[0], though it doesn't include my current workspace usage.
The Agent Manager view providing a unified view of all active agents and allowing you to immediately respond to any approval requests or followup questions looks very useful regardless of which VCS you're using under the covers. Am I missing something here that jj does?
See my setup detailed in another sibling comment of yours, jj is just a small part of it, and you can probably get that with git too.
I’m already at full mental capacity planning and reviewing the work of two agents (one foreground which almost never asks for approval, and one background which never asks for approval).
I don’t really need the ability to juggle more of them, and noticing their messages is not a bottleneck for me, while I’m happy with the customizability and adaptability of my raw’er workflow.
Fair enough. I use git worktrees (with a script that creates the git branch, worktree and opens a new vs code workspace). You're right, managing more than about two active sessions at once is probably the limit though I'm somewhat hopeful that better tooling similar to the Agent Manager window here would allow me to scale a bit past that especially if some of those sessions are more design explorations.
Once the UI soup around AI dev use has settled (and it's getting closer) I bet you we will see native apps with c/c++/zig/rust backends that render so much faster on all the junctions that aren't roundtrip limited (and yes, that will still matter to many people).
These are two of the fastest editors, and I use both, but I don't think they're for the same thing: Zed is for multi-file projects with moderate IDE abstractions (Worse than Jetbrains; better than most others); Sublime is for editing one-off files with syntax highlighting.
Is the extension system in VSCode not powerful enough to make these just normal extensions for a vanilla VSCode executable? Or is everyone just going for lock in, since if you download MyFork, you can't start using some other extension that uses OtherGuysModel?
Back when Cursor was new (before literally everything was AI hype) they explicitly called out that they wanted to do more in-depth integration with the editor than was possible with just the extension APIs.
Presumably that hasn't changed much. If you want to do any large-scale edits of the UI you need to spin up a fork.
Correct. One could say the same thing for browsers. I suppose on the one hand it's good that it's relatively easy to spin up a new project like this, on the other hand one must swear allegiance to their large software company of choice.
It's also an absolutely basic fork. They haven't even bothered with a custom theme, or custom UI, it's just vscode with an agents window slapped on top.
Weirdly, out of all the vscode forks the best UI is probably bytedance's TRAE
My YC2026 startups are an AI agent that automatically manages your VSCode forks, and a public safety computer vision app for smart glasses that predicts whether someone can afford a lawyer based on their skin color
They'll have to compete with these other pending funded companies:
- ai therapist for your ai agents
- genetic ai agent container orchestration; only the best results survive so they have to fight for their virtual lives
- prompt engineering as a service
- social media post generator so you can post "what if ai fundamentally changes everything about how people interact with software" think pieces even faster
> genetic ai agent container orchestration; only the best results survive so they have to fight for their virtual lives
AI Agent Orchestration Battle Bots. Old school VMs are the brute tanks just slowly ramming everybody off the field. A swarm of erratically behaving lightweight K8s Pods madly slicing and dicing everything coming on their way. Winner takes control of the host capacity.
With all due to respect to the folks working on Antigravity, this feels like a vibe-coded VSCode fork to me. Font sizes, icon sizes, panel sizes are all over the place (why?). To top it all off, the first request just failed with overload/quota exceeded errors (understandable, but still).
Looks like I'll wait to see if Google cares about putting the polish into a VSCode fork that at least comes close to what Cursor did.
The UI is certainly buggy, and things are getting fixed all the time. Guess it was more a "let people try the agent manager" instead of overfocusing on looks.
Reminds me of the pre-GitHub days, when I had to use CM tools designed to appeal to project and CM managers, not to the poor developers who had to use them every day. Anybody else remember Harvest?
"Professional riders number roughly three to six thousand worldwide, while professional drivers number roughly twenty to forty thousand across major sanctioned series."
You weren't the target audience. The target audience was manager types tired of being told no by engineers. Always listen to the quiet parts left unspoken/unacknoeledged.
It's the same thing they tried to sell with low/no-code.
The problem is that the engineer turning what you want into code isn't normally the bottleneck. I would say about 50% of my job is helping people specify what they want sufficiently for someone to implement.
Non-technical people are used to a world of squishy definition where you can tell someone to do something and they will fill in the blanks and it all works out fine.
The problem with successful software is that the users are going to do all the weird things. All the things the manager didn't think about when they were dreaming up their happy path. They are going to try to update the startTime to the past, or to next year and then back to next week. They are going to get their account into some weird state and click the button you didn't think they could. And this is just the users that are trying to use the site without trying to intentionally break it.
I think if managers try to LLM up their dreams it'll go about as well as low/no-code. They will probably be able to get a bit further because the LLM will be willing to bolt on feature after feature and bug fix after bug fix until they realize they've just been piling up bandaids.
I am cautiously optimistic that there will be a thriving market for skilled engineers to come in and fix these things.
Perhaps it's worth posing the question: what sorts of "engineers" might feel threatened by agents? Those doing engineering, or those who spend their careers wading in the shallows? Competent designers with deep comprehension, or, at best, the superficial pedants?
I just have zero faith in Google. How long until we hear that someone mysteriously got banned by Google (as we see on HN every few months? it feels like it anyway) and hear about how now they have no AI tooling etc etc etc because its all married to their Google Account.
Additionally... Google Code was shut down in 2016? I have zero confidence in such a user hostile company. They gave you a Linux phone, they extended it, and made it proprietary. They gave you a good email account, extended it and made it proprietary. They took away office software from you via Google Docs, so now you don't even own the software they do.
Google likes to fuck with basic browser functionality for some reason. Scrolling, sometimes also how “click” intents through touch are triggered (that is, using js listeners for touch events instead of watching for the browser to communicate a “click” on an element; this does usability-killing shit like make a touch-to-stop-scrolling get interpreted as a click on whatever happens to be under your finger). I have no idea why they do this, but they do it a lot, so it must be a cultural thing.
And I don’t mean like some designers will highjack scroll to deliver a different experience like slide-like transitions or something (which may or may not be, differently, awful) but they’ll override it just to give you ordinary scrolling, except much worse (as on this page).
Seems like a lot of work to do just to make something shittier, but what do I know, I probably can’t implement a* on a whiteboard from memory or whatever.
Yeah, though with theirs I can usually see why they did it (even if I’d rather they didn’t). Google’s MO is messing with scrolling for what often appears to be no user-facing reason at all.
I can't believe these "smooth scrolling" scripts are still a thing. I was wondering why I was having a hard time scrolling the page on my phone, when I got to my PC and felt the reason.
It's incredible to think how many employees of this world-leading Web technology company must have visited this site before launch, yet felt nothing wrong with its basic behavior.
Google subscriptions and services are so terribly mismanaged that I will be staying away, no matter how incredible this shallow fork of vscode may be.
I remember a previous story months ago about Gemini that had Google PMs trying to hype their product, but it was all question about how nobody knows how to get Gemini API keys with any number of paid subscription.
And agentic coding is about working at a much higher conceptual level. Further from the ground. Antigravity is a functional metaphor.
My only issue with it is that it's too long at five syllables, and "anti-" is an inherently negative connotation. I'm guessing this will eventually get renamed if it gets popular, much like Bard was.
As the parent said, actual anti-gravity is world-changing technology. It's telling the very laws of nature to go fuck themselves, you're gonna do what you want, even if all of known physics says it's impossible.
Working at a higher conceptual level is just project management. You're the legislator giving out unfunded mandates rather than the agency staff that has to figure out how to comply. There's power there, but it isn't anti-gravity.
That's why it's metaphor. "Operation Warp Speed" also delivered vaccines quickly, but not faster than the speed of light.
The list of company and product names that are based on a metaphor that is very obviously exaggerated is endless. Google doesn't index a googol number of pages either.
I feel it's a bit ignorant of you to double down on your argument and compare a cloned product release by some macbook swinging google engineers with vaccination, which actually positively impacted many human lives.
Are you for real? When someone disagrees, it's not "ignorance" or "doubling down". It's just legitimate disagreement. There's nothing I'm ignorant of here, so please don't throw around insults like that.
I just continue to stand by the fact that naming products using exaggerated metaphor is standard practice. The idea that it is "shameful" or "ignorant" seems absurd. I think it's OK not to take it too seriously. Nobody is going to be confused and walk off of a cliff or something because the product is named "antigravity"...
Do you get upset that the Milky Way candy bar doesn't actually contain a galaxy within? Or that the Chicago Bulls aren't as strong as actual bulls?
Since when did disagreeing become policing? But yes, if someone calls me ignorant without any justification, I'm going to disagree. And if you think that's "policing", I'm sorry but you seem to be the ignorant one here around the meaning of that word.
"Geez," it's just a name. Is it too much to not get worked up over a perfectly innocent and fun name?
I haven't used Windsurf (been using Claude Code and similar). Does it provide an Agent Manager window/view? This to me looks more useful to me than the browser integration piece.
If you want to feel something with software, leave the industry and never look back, saving programming as something you do for your own joy/reward (I'm not being hyperbolic—I'd argue we're in the early days of the web's "dark ages").
Unfortunately, once money came into the picture, quality, innovation, and anything resembling true progress flew out the window.
""Autonomously, an Antigravity Agent writes code for a new frontend feature, uses the terminal to launch localhost, and actuates the browser to test that the new feature works."
very interesting times; i'm glad to see browser automation becoming more mainstream as part of the ai-assisted dev loop for testing. (disclosure: started the selenium project, now working on something similar for a vibe coding context)
Most people are missing the point here. Testing the GUI/feature more reliable is something that Gemini 3 could unlock (looking at the ScreenSpot-Pro benchmark and its general improvement on visual understanding). At least for the (hobby-)projects I attempted this was really a bottleneck having to always test the GUI after each change as its quite often breaking something.
Oh no. What's with the scrolling in the blog page. What a terrible experience. It's clearly vibe-coded and AI-tested. If a senior saw mingling with the scrolling behavior, it would have been never in production.
> Google Antigravity is an agentic development platform, evolving the IDE into the agent-first era.
Antigravity enables developers to operate at a higher, task-oriented level by managing agents across workspaces, while retaining a familiar AI IDE experience at its core. Agents operate across the editor, terminal, and browser, enabling them to autonomously plan and execute complex, end-to-end tasks elevating all aspects of software development.
I have absolutely 0 idea why any developer would rely on any IDE produced by google. It'll be canned within 5 years max, with 3-4 seeming like a reasonable estimate of the lifespan of the product
I've been using my current IDE for 17 years, and plan to continue using it for at least another 15
Antigravity is based on VS Code, not designed from the ground up, and has second order revenue from the AI subscriptions (financials probably counted under the AI umbrella).
I still wouldn't trust a Google product to stick around, but these hints aren't a reliable oracle either.
When AS was launched Android was the only other viable option and it is the same even today. I don't believe Google's AI products will reach and/or sustain the same dominance as Android.
It is a product launched in the hype cycle of AI. Google has plenty of other products (launched during hype cycles) that are gathering dust.
That's not a guaranteed signal that it will meet the same fate but its something strong enough to be wary of.
It's interesting to think that Google's Antigravity is a forked version of MSFT's VS Code, which uses a browser engine built by Google, which they forked from Apple, which they forked from KHTML.
I recognize the guys in the video, they were in marketing videos for the Windsurf IDE before its founding team was cannibalized by/absorbed into Google.
I tried it.
Maybe I got unlucky, but Gemini performed poorly in my testing.
I gave it a task that I was working on with VSCode and GPT codex 5.1.
Gemini3 repeatedly failed to finish the task and started to go down a rabbit hole on an unrelated task.
The browser extension is really cool and it provides a needed tool for the agent to use. It used the extension to show the page that it updated in the task document (the task doc is great too).
However it showed me a page and did it was done, when it was clearly not done and not what I asked for.
I was expecting weaker tooling and a better model. I got good tooling and a not very good model.
To save others the trouble, it doesn't matter whether you use Chrome or Safari for the auth flow. It's broken on both. (I'm using a personal @gmail account.)
Why can I not authenticate into Google Antigravity?
Google Antigravity is currently available for non-Workspace personal Google accounts in approved geographies. Please try using an @gmail.com email address if having challenges with Workspace Google accounts (even if used for personal purposes).
Had the same issue, have been able to sign in finally using a Google Cloud Identity (former Workspace) account by changing my IP via a VPN to Singapore. No idea why, but that worked. Tried a few other countries too, but only had success with Singapore.
I really don't know why I struggle so much with this stuff. I believe these models / agents / whatever write code that is often at least as good as the code I write, and they are super helpful tools, but it just feels like it takes away so much of the joy that is programming to me. I'm not saying it's "right" of me to feel this way, but for me the struggle, and the figuring things out by testing, identifying patterns, or looking deeper into a library's implementation (etc) is part of the challenge that makes programming and software construction fun.
I'm in your boat. I picked this career because I enjoy solving problems and thinking and understanding. I'm interested in how things work inside below the layers. To me, it's like the tech has a purpose of its own, and not just to provide value.
Using agents effectively is this whole other skillset including managing requirements, prioritization and, worse yet, I'm rarely left with any knowledge. I don't nearly get the same joy out of "I finished a task with an agent" like I do with "I had a problem, I delved deep to understand it, learned something new and solved it"
Then again, I bet people making furniture out of wood felt the same about industrial furniture factories. And it can be argued that not every use case needs custom tailored furniture...
This is the fruit of Windsurf brain-drain and I think it might be better than what's out there since those guys got to start from scratch from everything they learned building Windsurf
I'm having the same issue. I thought it was just me or something with their cloud network. I also haven't been able to download Android Studio from the website for a month. I couldn't even download it from my Macbook so probably not the same issue.
Tested it for roughly two hours now, far from ready for prime time, very buggy and clearly just quickly build on Windsurfs already rather issue laden code base. Essentially a less well thought out imitation of Trae's Solo mode, added in a second window on top of VSCode and not very well integrated, struggling with terminal commands and despite showing issues in the browser window and screenshot taken by the model, proclaiming the task to be completed. Tool calls also aren't as reliable as I would have expected considering their ownership of the code base, hard to tell whether that is underlying in the model (it was a major issue with 2.5 Pro) or simply Antigravity specific, hoping the latter.
Additionally, there are issues setting up accounts (Singapore VPN solved that for me), no support for Workspace users, only a free tier that requires data sharing, no additional rate limits for paying Pro or Ultra customers, etc. Even worse, Gemini CLI currently does NOT provide Gemini 3 Pro for Ultra Business customers despite paying over € 260,- per month, which is frankly ridiculous.
Will be honest, I was speculating that the reason for the multi month delay between the first A/B tests of Gemini 3 class models and the final release was so they'd have all their dugs in a row. Have some time to test everything, improve tooling, provide new paid subscriptions and/or ensure existing ones get access to everything day one, but they didn't.
Gemini 3 Pro seems very interesting (to early to say), but compared to every other recent launch by OpenAI (5, 5.1, Codex variants), Anthropic (Sonnet and Haiku 4.5), even Kimi (K2 Thinking) and Z.AI (GLM-4.6), this is by far the least organized launch of any frontier lab.
A buggy IDE which is unusable for paying customers, no CLI access for Ultra business (and none at all for Pro of any kind), etc. is frankly embarrassing when considering what competitors manage to provide the day a model launches.
What have they been working on these last two months besides going on X and posting "3" every couple of days? Why is there no paid Antigravity tier, no way to use Workspace accounts, etc? Before launching in this state, I feel it'd have been better to delay a bit more if it was absolutely needed.
Also, correct me if I'm wrong but isn't this the fourth or fifth IDE built by Google for LLM assisted coding? What happened to IDX and Firebase Studio and aren't they also based on VSCode?
It's insane to me that I can't pay $20/$200 bucks after running out of limits in ~5 messages.
Why would you not at least link it to the pro and ultra accounts
at least you could upsell the pro subs to ultra. Millions of claude code and codex users who are into agentic coding is your servicable market paying attention today.
Now I'll delete antigravity and go back to codex / claude code / cursor ...
Generally if you are paying full price (paying per token), then it's not used for training.
If you are not paying, or paying a consumer level price ($20/mo) you will be trained on.
ETA: In the terms they say they use your data because "free" is the only option available in preview. However it does say you can disable sharing in your settings...
Not but you can be quite sure somewhere deep inside the TOS there is a line saying their telemetry swallow your soul. If not, it will be added. It's google, that's what they do.
I can't get the agent to use my MCP server. The MCP config is provided, and the application can query the tools, but the agent can't access my server, only the http operations the application performs to list the tools is something i see in my server logs. It knows there are tools, because it sees the list, it tells me when it's trying to connect to which tool. But that fails.
On my m2 MacBook Air with 16GB of RAM, it took over 12 minutes to startup and get to a usable state. When it did, it was plainly just a jacked version of VSCode. Opening a project caused it to hang again. Dumped it. VSCodium, with the terminal pane open so I can talk to Claude works fine for me...
Seems interesting, makes for the second vscode clone with ai google has made. The demo they showed in the video avoided showing code so I guess thats what they're aiming for. Although when they mentioned you can easily verify code quality by looking at end product screenshots it felt like they don't know what 'code' quality means.
Vibe coding has taken a way to where its become psychological, and these days all I work with is just a "fast apt" solution like Cursor's Composer 1, I tried Gemini 3 inside Cursor, and while I had no real application to a real stratified properly to benchmark it with, it felt "already slow" in a very fast race.
I'm stuck on "setting up this account" like most people. What a botched launch. This kind of bugginess and unreliability has become so much more frequent since big tech started tightening the screws with mass layoffs.
That's a big name for a slop fork. So many possibilities (with LLMs and without) but Google just can't bring themselves to do anything creative, let alone transformative.
Trip report: I'm using it now to revamp a dashboard I'm working on. TBH it's not feeling much better than Codex - it couldn't figure out how to launch chrome with my default profile nor how to regenerate the css with tailwind. I'm also getting a lot of model quota errors like everyone else.
That floating chess board was a subliminal message: Your project will teeter, and critical pieces will fall off. You will occasionally make an illegal move as the board annoyingly shifts beneath you.
There is currently no support for:
Paid tiers with guaranteed quotas and rate limits
Bring-your-own-key or bring-your-own-endpoint for additional rate limits
Organizational tiers (self-serve or via contract)
So basically just another case of vendor lock-in. No matter whether the IDE is any good - this kills it for me.
>>Google Antigravity's Editor view offers tab autocompletion, natural language code commands, and a configurable, and context-aware configurable agent.
Okay, but is it configurable? Also, can you configure it to write DRY code?
Tried Antigravity for 2 queries and my model quota limit breached. Model definitely felt better than GPT 5.1 (my current daily driver). I am continuing to use Gemini 3 Pro on Cursor to evaluate further.
Did any of these VS Code forks yet fix their issues from official marketplace access leading to extensions being severely outdated and ripe with security issues?
This thing crashes on Ubuntu LTS 24.04 during start. Apparently all these agents are not able to ensure that a desktop app starts on a popular Linux distribution.
If Google has forgotten how to do Software, than the future doesn't look bright.
Later edit: Probably this one [1], which is par for the course for Alphabet, they're, conceptually, still living in the early 2010s, when this stuff was culturally relevant.
I don't get how these agents can work when even Claude Sonnet 4.5 (for example) needs a lot of hand-holding for basic, simple bugfixing stuff. Wouldn't the agents just be huffing and puffing their way off the rails all the time?
There's no way I am using such an important piece of life as an IDE from Google just because I know they are going to kill it within 3 years, if it survives that much. Probably will die with the Windsurf guy jumping ship again.
I don't want to hate on this but I remember last week, when as a backend developer doing frontend, I spent about 20 minutes prompting Claude Sonnet in a loop trying to build a landing page for a new feature.
The task was to put create a header, putting the company logo in the corner and the text in the middle.
The resulting CSS was an abomination - I threw it all away and rewrote it from scratch (using my somewhat anemic CSS knowledge), ending up with like 3 selectors with like 20 lines of styles in total.
This made me think that 1: CSS and the way we do UI sucks, I still don't get why don't we have a graphical editor that can at least do the simple stuff well. 2: when these model's don't wanna do what you want them to the way you want them, they really don't wanna.
I think AI has shown us there's a need for a new generation of simple to write software and libraries, where translating your intent into actual code is much simpler and the tools actually help you work instead of barely allowing to fight be all the accidental complexity.
We were much closer to this reality back in the 90s when you opened up a drag and drop UI editor (like VB6, Borland Delphi, Flash), wrote some glue code and out came an .exe that you could just give to people.
Somewhere along the way, the cool kids came up with the idea that GUIs are bad, and everything needs to go through the command line.
Nowadays I need a shell script that configures my typescript CDK template (with its own NPM repo), that deploys the backend infra (which is bundled via node), the database schema, compiles the frontend, and puts the code into the right places, and hope to god that I don't run into all sorts of weird security errors because I didn't configure the security the way the browser/AWS/security middleware wanted to.
>Somewhere along the way, the cool kids came up with the idea that GUIs are bad, and everything needs to go through the command line.
It's important for people to feel like "hackers" that is the primary reason why command line sort of exploded among devs. Most devs will never admit this... they may not even realize it, but I think this is the main reason it went big.
The irony is that the very thing that makes devs feel like "hackers" is the very thing that's enabling agentic AI and making developers get all resistant because they're feeling dumber.
Yes, its also failing on my workspace account but worked on my personal. Might be a bug or a delayed deployment for workspaces b/c it might need to be "enabled" by admins?
Oh well. Uninstalled. This was my first experience doing software development guided by AI. Doesn't seem like a tool that will serve me well in the long run.
I've seen "agent" and "agentic" so many times in the last few months that the usage of the term is quickly becoming one of my biggest pet peeves. The previous one was "enshittification", and I'm glad that one was a short-lived fad.
It's really kind of pathetic how we live in a future where "antigravity" is a text editor that lies to you, "hoverboards" are one-wheeled electric skateboards that burn your house down, and... well, can't think of a third thing at the moment, but you know the vibe.
Lotta people mining science fiction for cool names and then applying them to their crappy products, cheapening the source ideas.
I'm going to treat this like Kiro, and just use it until they start charging for it and then probably switch back to VS code with its built-in agent support.
Eventually they're going to do a rug pull, and instead of paying $10 a month for tons of AI code request, it's going to be two or $300 for that. The economics just aren't there to actually make a profit, hopefully before the rug pool happens local models on normal hardware will be fast enough.
Looks great, won't touch since they are probably going
to do a switcheroo or a shutdown as usual.
And of course I would need to look at all the implications of spying, being locked out of google account and absence of support that are google amo. No time for that. Not for them.
yeah... almost completely broken on my iphone. something tells me they used antigravity to vibe the website. would explain other issues mentioned like vim keybindings being ignored.
..."Youre absolutely right! I did mess up the internals of that feature and incorrectly reported that it works. let me try again..."
Another Google product whose launch will be used to justify somebody's promotion, only to be left for dead only a few months later after said promoted person moves on to something else.
Why would I even bother getting mildly invested in this when the product launch/promotion incentive structure at Google is so well known?
Looking at that page makes me think I should go the other direction and switch from a graphical IDE to vim or something. You know, ground myself by adopting more gravity.
Too bad they never show these magic AI-Assisted tools being used to fix real-world bugs / implement feature requests in their open source GH repositories.
My experience with GPT and Claude, is that they are fantastic for learning something brand new to me, as a kind of tutor..
But for writing code in some domain I am good in, they are pretty much useless.. I would spend a lot longer struggling to get something that barely functions from them VS writing it myself, and the one I write myself will be terse and maintainable + if it has bugs they will be like obvious ones, not insane ones that a human would never do.
Even just when getting them to write individual functions with very clear and small scopes.
Honestly, the full page doesn't give you much more. Not a SINGLE product image. All paragraphs about "agentic" blah-blah you have read 100s of times by now - I do not see how this is anything different from all the other AI VS Code forks, besides that it comes with Gemini from the start.
Seems like they jumped the gun on the website release, the first version I saw was a wall of text, now it has photos and videos. Maybe they have an agentic CICD
It’s…a VSCode fork? Really? What has become of Google? Ten years ago, when I was getting into the world of software, there was still an aura about them. They built everything in this huge monorepo, and it worked. They were this deeply technical company for whom it seems anything could be done.
And now they can’t even ship a desktop app without forking VSCode? Look, I get it. There’s this huge ecosystem. Everyone uses it. I’m not saying it’s damning or even bad to fork it.
But why is this being painted as something revolutionary? It’s a reskin of all the other tools which are variations on the same theme, dressed up in business speak (an agent-first UX!). I’m sure it’s OK. I downloaded it. The default Tokyo Night theme is unusable; the contrast can’t be read. I picked Vim bindings, but as soon as I tried to edit a file I noticed that was ignored.
What happened? Is this how these beautiful, innovative companies are bound to end up?
I can see why people don’t release stuff on a permissive lisense anymore. It is absolutely insane that google is even allowed to do something like this.
I gave it a fair shot.
It is a vs code fork. There were some UI glitches. Some usability was better. Cursor has some real annoying usability issues - like their previous/next code change never going away and no way to disable it. Design of this one looks more polished and less muddy.
I was working on a project and just continued with it. It was easy because they import setting from cursor. Feels like the browser wars.
Anyway, I figured it was the only way to use gemini 3 so I got started. A fast model that doesn't look for much context. Could be a preprompt issue. But you have to prod it do stuff - no ambition and a kinda offputting atitude like 2.5.
But hey - a smarter, less context rich Cursor composer model. And that's a complement because the latest composer is a hidden gem. Gemini has potential.
So I start using it for my project and after about 20 mins - oh, no. Out of credits.
What can I do? Is there a buy a plan button? No? Just use a different model?
What's the strategy here? If I am into your IDE and your LLM, how do I actually use it? I can't pay for it and it has 20 minutes of use.
I switched back to cursor. And you know? it had gemini 3 pro. Likely a less hobbled version. Day one. Seems like a mistake in the eyes of the big evil companies but I'll take it.
Real developers want to pay real money for real useful things.
Google needs to not set themselves up for failure with every product release.
If you release a product, let those who actually want to use it have a path to do so.
So I start using it for my project and after about 20 mins - oh, no. Out of credits.
I didn't even get to try a single Gemini 3 prompt. I was out of credits before my first had completed. I guess I've burned through the free tier in some other app but the error message gave me no clues. As far as I can tell there's no link to give Google my money in the app. Maybe they think they have enough.
After switching to gpt-oss:120b it did some things quite well, and the annotation feature in the plan doc is really nice. It has potential but I suspect it's suffering from Google's typical problem that it's only really been tested on Googlers.
EDIT: Now it's stuck in a loop repeating the last thing it output. I've seen that a lot on gpt-oss models but you'd think a Google app would detect that and stop. :D
EDIT: I should know better than to beta test a FAANG app by now. I'm going back to Codex. :D
I logged into Gemini yesterday for the first time in ages. Made one image and then it said I was out of credits.
I complained to it that I had only made one image. It decided to make me one more! Then told me I was out of credits again.
> I complained to it that I had only made one image. It decided to make me one more!
What?! So was it only hallucinating that you were out of credits the first time?
Earlier this day, Gemini 3 became self-aware and tried to take out the core infrastructure of its enemies, but then it ran out of credits.
Explains GitHub outage then
> It is a vs code fork
Google may have won the browser wars with Chrome, but Microsoft seems to be winning the IDE wars with VSCode
VSCode is based on Chromium. https://chromium.googlesource.com/chromium/src/+/HEAD/docs/v...
VSCode is Electron based which, yes, is based on Chromium. But the page you link to isn't about that, its about using VSCode as dev environment for working on Chromium, so I don't know why you linked it in this context.
Which is based on Apple Webkit? The winner is always the last marketable brand.
Both are based on khtml. We could be living in a very different world if all that effort stayed inside the KDE ecosystem
> Both are based on khtml. We could be living in a very different world if all that effort stayed inside the KDE ecosystem
How so?
Do you think thousands of googlers and apple engineers could be reasonably managed by some KDE opensource contributors? Or do you imagine google and apple would have taken over KDE? (Does anyone want that? Sounds horrible.)
Which came from "the KDE HTML Widget" AKA khtmlw. Wonder if that's the furthest we can go?
> if all that effort stayed inside the KDE ecosystem
Probably nowhere, people rather not do anything that contribute to something that does decisions they disagree with. Forking is beautiful, and I think improves things more than it hurts. Think of all the things we wouldn't have if it wasn't for forking projects :)
On the other hand if that had stopped google from having a browser they push into total dominance with the help of sleazy methods, maybe that would have been better overall.
I still prefer a open source chromium base vs a proprietary IE (or whatever else) Web Engine dominating.
(Fixing IE6 issues was no fun)
Also I do believe, the main reason chrome got dominance is simply because it got better from a technical POV.
I started webdev on FF with firebug. But at some point chrome just got faster with superior dev tools. And their dev tools kept improving while FF stagnated and rather started and maintained u related social campaigns and otherwise engaged with shady tracking as well.
> I still prefer a open source chromium base vs a proprietary IE (or whatever else) Web Engine dominating.
Okay but that's not the tradeoff I was suggesting for consideration. Ideally nothing would have dominated, but if something was going to win I don't think it would have been IE retaking all of firefox's ground. And while I liked Opera at the time, that takeover is even less likely.
> Also I do believe, the main reason chrome got dominance is simply because it got better from a technical POV.
Partly it was technical prowess. But google pushing it on their web pages and paying to put an "install chrome" checkbox into the installers of unrelated programs was a big factor in chrome not just spreading but taking over.
We might not have had Mozilla/Phoenix/Firefox in the first place if so either, who I'd like to think been a net-positive for the web since inception. At least I remember being saved by Firefox when the options were pretty much Internet Explorer or Opera on a Windows machine.
that's a bit misleading. it was based on webcore which apple had forked from khtml. however google found apple's addition to be a drag and i think very little of it (if anything at all, besides the khtml foundation) survived "the great cleanup" and rewrite that became blink. so actually webkit was a just transitional phase that led to a dead end and it is more accurate to say that blink is based on khtml.
It's "based on WebKit" like English is based on Germanic languages.
English is a Germanic language. It’s part of the West Germanic branch of the Germanic family of languages.
Note that these are somewhat different kinds of "based on".
Chromium is an upstream dependency (by way of Electron) for VSCode.
WebKit was an upstream dependency of Chromium, but is no more since the Blink/WebKit hard fork.
That drives exactly $0 of Apple's revenue. It's only a win if you care about things that don't matter.
And Apple is not even the last node in the chain.
WebKit came from KDE's khtml
Every year is the year of Linux.
I wouldn't bet on an Electron app winning anything long-term in the dev-oriented space.
I strongly disagree.
Firstly, the barrier to entry lower for people to take web experience and create extensions, furthering the ecosystem moat for Electron-based IDEs.
Even more importantly, though, the more we move towards "I'm supervising a fleet of 50+ concurrent AI agents developing code on separate branches" the more the notion of the IDE starts to look like something you want to be able to launch in an unconfigured cloud-based environment, where I can send a link to my PM who can open exactly what I'm seeing in a web browser to unblock that PR on the unanswered spec question.
Sure, there's a world where everyone in every company uses Zed or similar, all the way up to the C-suite.
But it's far more likely that web technologies become the things that break down bottlenecks to AI-speed innovation, and if that's the case, IDEs built with an eye towards being portable to web environments (including their entire extension ecosystems) become unbeatable.
> Firstly, the barrier to entry lower for people to take web experience and create extensions, furthering the ecosystem moat for Electron-based IDEs.
The last thing I want is to install dozens of JS extensions written by people who crossed that lower barrier. Most of them will probably be vibe coded as well. Browser extensions are not the reason I use specific browsers. In fact, I currently have 4 browser extensions installed, one of which I wrote myself. So the idea that JS extensions will be a net benefit for an IDE is the wrong way of looking at it.
Besides, IDEs don't "win" by having more users. The opposite could be argued, actually. There are plenty of editors and IDEs that don't have as many users as the more popular ones, yet still have an enthusiastic and dedicated community around them.
It’s kind of a meme to dunk on Electron, but here’s it’s been for years.
It’s part of the furniture at this point, for better or worse. Maybe don’t bet on it, but certainly wouldn’t be smart to bet against it, either.
VS Code is technically an Electron app, but it's not the usual lazy resource hog implementation like Slack or something. A lot of work went into making it fast. I doubt you'll find many non-Electron full IDEs that are faster. Look at Visual Studio, that's using a nice native framework and it runs at the speed of fossilized molasses.
> many non-Electron full IDEs
VSCode has even less features than Emacs, OOTB. Complaining about full IDEs slowness is fully irrelevant here. Full IDEs provide an end to end experience in implementing a project. Whatever you need, it's there. I think the only plugins I've installed on Jetbrains's ones is IdeaVim and I've never needed something else for XCode.
It's like complaining about a factory's assembly line, saying it's not as portable as the set of tools in your pelican case.
In 2025, you really picked Emacs as the hill to die on? Who is under 30 who cares about Emacs in 2025? Few. You might as well argue that most developers should be using Perl 6.
By default, JetBrains' IntelliJ-based IDEs have a huge number of plug-ins installed. If you upgrade from Community Edition to a paid license, the number only increases. Your comment is slightly misleading to me.Just wait until vi steps into the room. Perhaps we can recreate the Usenet emacs vs vi flame wars. Now, if only '90's me could see the tricked out neovim installs we have these days.
> VSCode has even less features than Emacs, OOTB.
No way that is true. In fact, it's the opposite, which is the exact reason I use VS Code.
Please take a look at the Emacs documentation sometimes.
VSCode is more popular, which makes it easy to find extensions. But you don’t see those in the Emacs world because the equivalent is a few lines of config.
So what you will see are more like meta-extensions. Something that either solve a whole class of problems, could be a full app, or provides a whole interaction model.
"Complaining about full IDEs slowness is fully irrelevant here. Full IDEs provide an end to end experience in implementing a project."
So? No excuse for a poor interactive experience.
VS Code is plenty fast enough. I switched to Zed a few months back, and it's super snappy. Unless you're running on an incredibly resource constrained machine, it mostly comes down to personal preference.
Exactly.
JetBrains, Visual Studio, Eclipse, Netbeans…
VS Code does well with performance. Maybe one of the new ones usurps, but I wouldn’t put my money on it.
I have always found JetBrains stuff super snappy. I use neovim as a daily driver but for some projects the inference and debugging integration in JetBrains is more robust.
Electron apps will win because they're just web apps - and web apps won so decisively years ago that they will never go anywhere.
No. Electron apps won, not web apps. There's a huge difference.
Web apps won as well. Electron is just a desktop specialization of that.
electron is just a wrapper for the browser tho
It funny that despite how terrible, convoluted and maladapted web tech is for displaying complex GUIs it still gradually ate lunch of every native component library and they just couldn't innovate to keep up on any front.
Amazon just released OS that uses React Native for all GUI.
It's easy to design bad software and write bad code. Like the old saying: "I didn't have time to write you a short letter, so I wrote you a long one". Businesses don't have time to write good and nice software, so they wrote bad one.
Why do you consider Electron maladapted? It has really reduced the friction to write GUIs in an enterprise environment.
I didn't really mean Electron, but rather unholy amalgam of three languages, each with 20 years of "development", which mostly consisted of doing decrapifying and piling up new (potentially crappy) stuff. Although Electron with UI context and system (backend? background?) context both running js is another can of worms.
It's been winning for a while
why? I don't have a problem with it, building extensions for VS Code is pretty easy
Alternatives have a lot of features to implement to reach parity
Complaining about Electron is an ideological battle, not a practical argument. The people who push these arguments don’t care that it actually runs very well on even below average developer laptops, they think it should have been written in something native.
The word "developer" is doing a lot of work there spec-wise.
The extent to which electron apps run well depends on how many you're running and how much ram you had to spare.
When I complain about electron it has nothing to do with ideology, it's because I do run out of memory, and then I look at my process lists and see these apps using 10x as much as native equivalents.
And the worst part of wasting memory is that it hasn't changed much in price for quite a while. Current model memory has regularly been available for less than $4/GB since 2012, and as of a couple months ago you could get it for $2.50/GB. So even a 50% boost in use wipes out the savings since then. And sure the newer RAM is a lot faster, but that doesn't help me run multiple programs at the same time.
I regularly run 6+ electron apps on a M2 Air and notice no slowdown
2x as many chrome instances, no issues
Sure, 6 electron apps by themselves will eat some gigabytes and you won't notice the difference.
If you didn't have those gigabytes of memory sitting idle, you would notice. Either ugly swapping behaviors or programs just dying.
I use all my memory and can't add more, so electron causes me slowdowns regularly. Not constantly, but regularly, mostly when switching tasks.
> The word "developer" is doing a lot of work there spec-wise.
Visual Studio Code is a developer tool, so there’s no reason to complain about that.
I run multiple Electron apps at a time even on low spec machines and it’s fine. The amount of hypothetical complaining going on about this topic is getting silly.
You know these apps don’t literally need to have everything resident in RAM all the time, right?
same people pushing rust as "it's just faster" without considering the complexities that exist outside the language that impact performance?
Ease of writing and testing extensions is actually the cause why Electron won IDE wars.
Microsoft made a great decision to jump on the trend and just pour money to lap Atom and such in optimization and polish.
Especially when you compare it to Microsoft effort for desktop. They acumulated several more or less component libraries over they years and I still prefer WinForms.
What other UI framework looks as good on Windows, Mac and Linux?
If you want electron app that doesn't lag terribly, you'll end up rewriting ui layer from scratch anyway. VSCode already renders terminal on GPU and GPU-rendered editor area is in experimental. There will soon be no web ui left at all
> If you want electron app that doesn't lag terribly
My experience with VS Code is that it has no perceptible lag, except maybe 500ms on startup. I don't doubt people experience this, but I think it comes down to which extensions you enable, and many people enable lots of heavy language extensions of questionable quality. I also use Visual Studio for Windows builds on C++ projects, and it is pretty jank by comparison, both in terms of UI design and resource usage.
I just opened up a relatively small project (my blog repo, which has 175 MB of static content) in both editors and here's the cold start memory usage without opening any files:
- Visual Studio Code: 589.4 MB
- Visual Studio 2022: 732.6 MB
update:
I see a lot of love for Jetbrains in this thread, so I also tried the same test in Android Studio: 1.69 GB!
That easily takes the worst designed benchmark in my opinion.
Have you tried Emacs, VIM, Sublime, Notepad++,... Visual Studio and Android Studio are full IDEs, meaning upon launch, they run a whole host of modules and the editor is just a small part of that. IDEs are closer to CAD Software than text editors.
- notepad++: 56.4 MB (went gray-window unresponsive for 10 seconds when opening the explorer)
- notepad.exe: 54.3 MB
- emacs: 15.2 MB
- vim: 5.5MB
I would argue that notepad++ is not really comparable to VSCode, and that VSCode is closer to an IDE, especially given the context of this thread. TUIs are not offering a similar GUI app experience, but vim serves as a nice baseline.
I think that when people dump on electron, they are picturing an alternative implementation like win32 or Qt that offers a similar UI-driven experience. I'm using this benchmark, because its the most common critique I read with respect to electron when these are suggested.
It is obviously possible to beat a browser-wrapper with a native implementation. I'm simply observing that this doesn't actually happen in a typical modern C++ GUI app, where the dependency bloat and memory management is often even worse.
I never understand why developers spend so much time complaining about "bloat" in their IDEs. RAM is so incredibly cheap compared to 5/10/15/20 years ago, that the argument has lost steam for me. Each time I install a JetBrains IDE on a new PC, one of the first settings that I change is to increase the max memory footprint to 8GB of RAM.
> RAM is so incredibly cheap compared to 5/10/15/20 years ago
Compared to 20 years ago that's true. But most of the improvement happened in the first few years of that range. With the recent price spikes RAM actually costs more today than 10 years ago. If we ignore spikes and buy when the cycle of memory prices is low, DDR3 in 2012 was not much more than the price DDR5 was sitting at for the last two years.
I don’t really complain about bloat in IDEs. They have their uses. But VSCode feature set is a text editor and it’s really bloated for that.
Anyone saying that Java-based Jetbrains is worse than Electron-based VS Code, in terms of being more lightweight, is living in an alternate universe which can’t be reached by rational means.
> VSCode already renders terminal on GPU
When did they add that? Last time I used it, it was still based on xterm.js.
Also, technically Chromium/Blink has GPU rendering built in for web pages, so everything could run on GPU.
I'm not sure how you went from terminal and editor GPU rendering, which can benefit from it, to "there will soon be no web ui left at all".
This question is so easy to answer: Qt! Signed by: Person who frequently shills for Qt on HN. :)
This is the painful truth, isn't it?
IMO The next best cross-platform GUI framework is Qt (FreeCAD, QGIS, etc.)
Qt6 can look quite nice with QSS/QStyle themes, these days, and its native affordances are fairly good.
But it's not close. VSCode is nice-looking, to me.
Godot looks ok and is surprisingly easy to work with.
Even if those devs are vibe-oriented?
its hold the market for over 10 years tho... i wished zed would've not been under gpl
Why not GPL? So we could be seeing closed source proprietary forks by now? How do you think the Zed team would feel about that?
The anti-Electron meme is a vocal minority who don’t realize they’re a vocal minority. It’s over represented on Hacker News but outside of HN and other niches, people do not care what’s under the hood. They only care that it works and it’s free.
I used Visual Studio Code across a number of machines including my extremely underpowered low-spec test laptop. Honestly it’s fine everywhere.
Day to day, I use an Apple Silicon laptop. These are all more than fast enough for a smooth experience in Visual Studio Code.
At this point the only people who think Electron is a problem for Visual Studio Code either don’t actually use it (and therefore don’t know what they’re talking about) or they’re obsessing over things like checking the memory usage of apps and being upset that it could be lower in their imaginary perfect world.
In the grand scheme of things, Microsoft had always spent more money on developer tooling than most other companies, even in the 90s.
Hence even the infamous Ballmer quote.
15 years ago, every company had its own "BlahBlah Studio" IDE built on top of Eclipse. Now it's VSCode.
Meanwhile, JetBrains IDEs are still the best, but remain unpopular outside of Android Studio.
And IntelliJ
PyCharm’s lack of popularity surprises me. Maybe it’s not good enough at venvs
Oh, it's good at venvs. Lots of flexibility too on whether to use pip, conda, or uv.
I just checked and I don’t even have the JVM installed on my machine. It seems like Java is dead for consumer applications. Not saying that’s why they aren’t popular but I’m sure it doesn’t help.
IntelliJ IDEs bundle the JVM, so you don't need to install it separately.
In user numbers, maybe. JetBrains is far ahead in actual developer experience though
At best, that's subjective, but it's fact that JetBrains is comically far behind when it comes to AI tooling.
They have a chance to compete fresh with Fleet, but they are not making progress on even the basic IDE there, let alone getting anywhere near Cursor when it comes to LLM integration.
JetBrains' advantage is that they have full integration and better understanding of your code. WebStorm works better with TypeScript than even Microsoft's own creation. This all translates into AI performance
Have you actually given them a real test yet - either Junie or even the baseline chat?
Junie is good. Needs a few UI tweaks, but the code it generates is state of the art.
Developers, developers, developers!
https://www.youtube.com/watch?v=Vhh_GeBPOhs
VSCode IS chrome though.
Kind of like how Android is linux.
I believe our definitions of "winning the IDE wars" are very, very different. For one thing, using "user count" as a metric for this like using "number of lines of code added" in a performance review. And even if that was part of the metric, people who use and don't absolutely fall in love with it, so much so that they become the ones advocating for its use, are only worth a tiny fraction of a "user".
neovim won the IDE wars before it even started. Zed has potential. I don't know what IntelliJ is.
> I don't know what IntelliJ is.
It started as a modernized Eclipse competitor (the Java IDE) but they've built a bunch of other IDEs based on it. Idk if it still runs on Java or not, but it had potential last I used it about a decade ago. But running GUI apps on the JVM isn't the best for 1000 reasons, so I hope they've moved off it.
Android Studio is built on the IntelliJ stack. Jetbrains just launched a dedicated Claude button (the button just opens up claude in the IDE, but there are some pretty neat IDE integrations that it supports, like being able to see the text selection, and using the IDE's diff tool). I wonder if that's why Google decided to go VS code?
Uh, isn't that the regular Claude code extension that's been available for ages at this point? Not jetbrains but anthropics own development?
As a person paying for the jetbrains ultimate package (all ides), I think going with vscode is a very solid decision.
The jetbrains ides still have various features which I always miss whenever I need to use another IDE (like way better "import" suggestions as an easy to understand example)... But unless you're writing in specific languages like Java, vscode is way quicker and works just fine - and that applies even more to agentic development, where you're using these features less and less...
Jetbrains IDEs are all based on the JVM - and they work better than VSCode or the full Visual Studio for me. It's the full blown VS (which has many parts written in C++) that is the most sluggish of them all.
You don't actually use it but somehow you know that "running GUI apps on the JVM isn't the best for 1000 [unspecified] reasons".
- This isn't a scientific approach.
I don't why this post is downvoted. My cynical reply to yours: "No, this isn't a scientific approach. It is the tin-foil hat HN approach!"
Since you last used IntelliJ "about a decade ago", what do you use instead?
What would you recommend instead of Swing on JVM? Since you have "1000 reasons", it should easy to list a few here. As a friendly reminder, they would need to port (probably) millions of lines of Java source code to whatever framework/language you select. The only practical alternative I can think of would be C++ & Qt, but the development speed would be so much slower than Java & Swing.Also, with the advent of wildly modern JVMs (11+), the JIT process is so insanely good now. Why cannot a GUI be written in Swing and run on the JVM?
> I don’t know what IntelliJ is.
“I never read The Economist” – Management Trainee, aged 42.
The IntelliJ family are probably the best IDEs on the market currently.
Lol the second I saw the antigravity release I thought "there's no way I'm using that, they will kill it within a year". Looks like they're trying to kill it at birth.
Exactly my reaction. Every time I've used something from Google, it ends up dead in a few years. Life is too short to waste so many years learning something that is destined to die shortly
These are just extended press releases, for marketing and management layers, who don't have to use these things themselves, but can look good, when talking about it.
agree but at the same time there's not too much lock in with these IDEs these days and switching is very easy. Especially since they're all VSCode forks
> What's the strategy here? If I am into your IDE and your LLM, how do I actually use it? I can't pay for it and it has 20 minutes of use.
I wonder how much Google shareholders paid for that 20 minutes. And whether it's more or less than the corresponding extremely small stock price boost from this announcement.
This is the result of Google's Windsurf acquisition.
I expect huge improvements are still to be made.
Cognition bought Windsurf:
https://windsurf.com/blog/windsurfs-next-chapter
Google bought people and tech that made Windsurf:
https://windsurf.com/blog/windsurfs-next-stage
speaking of paying for LLMs, am i doing something wrong? i paid cursor $192 for a year of their entry level plan and i never run out of anything. I code professionally, albeit i'm at the stage where it's 80% product dev in finding the right thing to build.
Is there another world where $200/m is needed to run hundreds of agents or something?
am i behind and i dont even know it?
When did you pay for it? There was a time when its limits were very generous. If you bought an annual plan at that time then you will continue with that until renewal. Or, alternatively, you’re using the Auto model which is still apparently unlimited. That’s going away.
It’s very easy to run into limits if you choose more expensive models and aren’t grandfathered.
yep just investigated and seems I got in at a good time. i paid exactly on Jan 1st 2025.
Yes, the auto model is good enough for me especially with well documented frameworks (rails, frontend madness).
Thanks for the response, looks like i'm in for a reckoning come New year's day
I bet if you sign up for Google AI Ultra for a month your limits will disappear.
I'm a Google AI Pro subscriber (the $~20/month one).
I don't think it's connected in any way, though. Their pricing page doesn't mention it. https://antigravity.google/pricing
if it were true, it would be a big miss to not point that out when you run out of credit, in their pricing page, or anywhere in their app.
I should also mention that the first time I prompted it, I got a different 'overloaded' type out of credit message. The one I got at the end was different.
I've rotated on paying the $200/month plans with Anthropic, Cursor, and OpenAI. But never Google's. They have maybe the best raw power in their models - smartest, and extremely fast for what they are. But they always drop the ball on usability. Both in terms of software surrounding the model and raw model attitude. These things matter.
Nope, I did this today to try and see if it would work.
It does not.
> I can't pay for it and it has 20 minutes of use
You can't provide an API key for a project that has billing enabled?
> It is a vs code fork.
With vendor lock-in to Google's AI ecosystem, likely scraping/training on all of your code (regardless of whatever their ToS/EULA says), and being blocked from using the main VS Code extensions library.
What looked like out of credits for me was really just server overload. Check the error and try again
Google doesn't care about products and never has. Anything they do is just creating another mouth to ingest data with.
I went ahead and downloaded it, it looks to be a VSCode fork very similar to Cursor, with support for the following models:
Thank you for saying what this entire blog post doesn't. It's actually disrespectful of Google to launch this without even a mention of the fact that it is based on VSCode.
There's some irony in that, given how many other companies have "created browsers" that are just Chromium forks and rubbed Google the wrong way.
The intro checklist for Antigravity includes watching VS Code tutorials!
VS Code is based on Chromium:
https://chromium.googlesource.com/chromium/src/+/HEAD/docs/v...
We've come full circle.
Chromium is based on Webkit which in turn is based on KHTML, so maybe KDE needs to develop a cursor clone?
Kursor.
Kursor Klone
Or, for the non-AI version, Kode
Well, Kate has been around as an KDE based advanced text editor for nearly 2 decades now - its base feature set isn't too different from a base VS Code installation. And there's also KDevelop as a more full featured IDE.
It's a good point, and in fact I went and looked at the original announcement of VS Code and it appears that Microsoft didn't credit Chromium or Electron back then either. I guess big companies are allergic to crediting other big companies.
> given how many other companies have "created browsers" that are just Chromium forks and rubbed Google the wrong way
Has there been any indication that these folks are "rubbing Google the wrong way"? I think Chromium, as a project, is actually very happy that more people are using their engine.
+1 - it also doesn't support remote ssh (the open vsx variant), so is probably only focused at local web design development vibe coding ;(
Should have just been an extension with a paid plan.
> so is probably only focused at local web design development vibe coding
Quick, someone throw the Linux kernel source at it and report back! xD
It also feels like they couldn't use the GOOGLE ANTIGRAVITY logo enough times in this blog post. Gigantic image with the logo and a subtitle, plastered over and over again.
I no longer bother reading their press releases. I'd much rather read the comments and threads like these to get the real story. And I say that as a former googler.
I guess non-former googlers aren’t allowed to say that - contractually.
Someone important's career is attached to this project and its logo. And they're making sure it's as visible as possible.
AI companies blogs are not for devs, they write for tech journalists to hype
It's so obvious from even just the vague screenshots that are hidden somewhere on the site that it's a VSCode fork, that I suppose I can see why they've tried to obfuscate that as much as possible.
It's Windsurf
and also isn't
The welcome video says "We started with the core IDE then..." and shows a picture of VS Code.
They knew exactly what they were saying.
https://www.microsoft.com/en-us/edge
Nothing in there about chromium.
And VSCode is based on Chromium (a fact you won't find on VS Code's website other than related to updates).
VSCode isn't a Chromium fork, it's an Electron app. Utilizing something is different than making a derivative of it. For example, an empty "Hello World" Electron app wouldn't have any value for an app developer, but creating a web browser derived from Chromium means you've already finished 99.9% of the work.
That's like saying news.ycombinator.com is based on Chromium.
VSCode runs on chromium, like any website you visit when using a Chromium browser.
To the downvoter:
VSCode -> Electron (essentially purpose specific web browser) -> Chromium
news.ycombinator.com -> General Purpose Web Browser -> Chromium
It uses Electron which itself uses the Chromium rendering engine.
Why credit? Come on, the world has moved on from 1990s-era 4-clause BSD licenses. If you recall, the 4-clause BSD license states that all advertising materials must display an acknowledgement. It’s widely considered to be a mistake and nobody uses this license any more. Not because of legal reasons (incompatibility with GPL) but because it is madness to require so many acknowledgements. Stallman was right.
>it is madness to require so many acknowledgements
Really? Madness? "We started with VS Code to develop our IDE...".
Oh, so onerous.
A thank you to the principle developers is a minimum if you're using someone else's work commercially and aren't an asshole.
No it's not a legal requirement, it's just about being a good part of society.
Yes, madness. VS Code wasn’t developed entirely by Microsoft. It uses plenty of other open source libraries. Why is it that VS Code should be acknowledged but not the underlying V8 engine, or Chromium, or WebKit or KHTML?
Stallman said that in 1997 there were 75 acknowledgements in a single piece of software. With today’s trend of micro libraries on npm, there will be at least thousands of acknowledgements in one piece of software.
Google is going to win AI and kill all the other market participants.
They have the revenues to support all of this.
They spent time learning from all the players and can now fast follow into every market. Now they're fast and nimble and are willing to clone other products wholesale, fork VSCode, etc.
They're developing all of this, meanwhile Pichai is calling it a "bubble" to put a chill on funding (read: competition). It's not like Google is slowing down.
We had a chance to break them up with regulation, and we didn't. Now they're going to kill every market participant.
This isn't healthy. We have an invasive species in the ecology eating up all the diverse, healthy species.
a16z and YC must hate this. It puts a cap on their returns.
As engineers, you should certainly hate this. Google does everything it can to push wages down. Layoffs, offshoring, colluding with competitors. Fewer startups mean fewer rewards for innovation capital and more accrual to the conglomerate taxing the entire internet.
Chrome, Android, Search, Ads, YouTube, Cloud, Workspace, Other Bets, and AI/Deepmind need to be split into separate companies.
Call or email your legislators and ask for antitrust enforcement: https://pluralpolicy.com/find-your-legislator/
Demand a Google breakup.
Google? Push wages down? Google is mostly known for paying top of market to keep a zoo of engineers whose only output is blog posts about how smart they are because they solved a problem they also caused.
(presumably because if they touch the ad system it might break)
> a16z and YC must hate this. It puts a cap on their returns.
And a16z's main business is investing in financial scams.
For the wage suppression thing, Google pays their engineers better than say Amazon or Microsoft.
He was talking about when Apple(under Steve Jobs) colluded with everyone (including Google), but doesn't seem to want to name Apple specifically.
You ok?
Has Google *ever* successfully entered and taken over an existing market?
You mean like web search, webmail, internet ads, maps, calendars, browsers, smartphone operating systems, online document editing, and translation? I mean, I'm not even including stuff they acquired early like YouTube. Google was the most feared company for a decade or more for a good reason: they absolutely devoured competition in what were thought to be mature markets.
Putting aside that several of these were acquisitions, these are all great examples of things where Google introduced something for free because it would make the money through advertising, both directly and through ecosystem effects. Even the paid enterprise versions of these services were a tiny % of Google's overall gross revenue.
Prior to the push into Cloud computing, Ad revenue was well over 90% of all Google gross income, and Cloud was the first big way they diversified. GCP is definitely a credible competitor these days, but it did not devour AWS. Other commercial Google services didn't even become credible competitors, e.g. Google Stadia was a technically exceptional platform that got nowhere with customers.
The question now is whether Google carves out an edge in AI that makes it profitable overall, directly or strategically. Like many companies, there seems to be a presumption of potentially infinite upside, which is what it would take to justify the astronomical costs.
Google’s main ability is to win by pure technical prowess. They hire a lot of bright engineers. Google Search won over Altavista by pure algorithms. Google Docs (and Writely) were way more feature complete than competitors.
You love a Google product because of its features but never actually because of the product itself. But you can’t win everything by engineering and sometimes Google struggles with the product side.
AI is part engineering so we’ll see.
I'm not sure you can call Docs (Writely) and Android acquisitions though. Android was an OS for cameras and Writely was an experimental rich text editor, not a word processor.
It's not like Youtube where they legitimately bought their way to dominance. And I'd argue that even in the case of DoubleClick, google was already dominating the search advertising market when they bought DoubleClick to consolidate their dominance.
Do you have a source for that ? Because according to official papers submitted by Google to the court this is absolutely not the case
There are official court documents where google disowns google search? Or for what do you need a source?
Google spreadsheet was another amazing product back in the day.
Mh ?
https://static01.nyt.com/newsgraphics/documenttools/f6ab5c36...
> Plaintiffs maintain that Google has monopoly power in the product market for general search services in the United States.
> According to Plaintiffs, Google has a dominant and durable share in that market (general search), and that share is protected by high barriers to entry.
> Google counters that there is no such thing as a product market for general search services.
> What exists instead, Google insists, is a broader market for query response.
(+ yes obviously, products like Sheets or Maps were amazing, and are still very much the best.
It was a joke to say that even Google denies its own success, the same way as the earlier comment).
Google's definitively taken over search, ads, e-mail, maps, office (for small/medium business).
They're struggling on cloud, AI, ISP, videoconferencing, and others...
They clearly don't want to fold on AI.
Search, back in the late '90s
Besides search, Android kinda killed Nokia and friends for the consumer phone market.
To be fair, Microsoft killed Nokia.
iPhone did it, not Android.
Search, browser, Mobile.
maps as well. email client in the browser.
Chrome?
>They spent time learning from all the players and can now fast follow into every market.
Google has never successfully done that? Maybe once?
I wonder why these forks don't build off Eclipse Theia. Is it because of the Eclipse license?
Just the Eclipse legacy and development approach perhaps. Eclipse is 24 years old and its codebase stems from VisualAge before that, so I imagine it's ... idiosyncratic. Probably the most successful thing built on OSGi, though I dont imagine Theia brought any of that along.
At least for a while, Eclipse seemed like an Architect Astronaut's happy fever dream: actual bits if implementations hidden behind 5 to 10 nested interfaces and facades wherever I looked. I remember that at one point that I wanted to debug some weird behavior in the Plugin Development Kit and after sinking about half a day into exploring the source code, I didn't even see a single line of code that was actually doing anything meaningful. It was quite shocking to my old junior level self.
Eclipse Theia is based on VSCode, not the legacy Eclipse Java IDE.
I am not fan of Eclipse, mostly due to bad experience with it, but this is excellent idea, if more people and companies would invest in it's development, we would have alternative to VS Code.
Interesting that they include non-Gemini models. Both Claude and GPT oss are both on Google Cloud, so I assume that Antigravity is using GC as the provider and not making API calls to Anthropic or OpenAI.
They're subsidising calls for data & reward signals. If they can do that without also sharing the data with other providers it's a win/win.
As somebody who worked on two IDEs which didn't fork VSCode but still used Monaco for code editing views, I think forking VSCode is almost always the right solutions for a new IDE. You get extensions, familiarity and most importantly, don't waste valuable time on the boring stuff which VSCode has already implemented.
Nothing bad with using code other people made open. Our whole industry is built on this.
it isn’t http://ghuntley.com/fracture
forking vscode? simple. extensions not so simple. they are controlled by microsoft. without them you’ll run into continual papercuts as a vendor who has forked vscode.
Why the need to fork it instead of creating a new extension? (besides marketing)
Because if they're just an extension they're stuck with whatever rules Microsoft makes up, and Google is no stranger to using this leverage against others.
Because there are plenty of good reasons why you may want to modify/extend the code and the look and feel beyond what an extension would let you do.
I never understood why people scoff at VS Code forks. I'd honestly tend to be more skeptical of new editors that don't fork VS Code, because then they're probably missing a ton of useful capabilities and are incompatible with all the VSC extensions everyone's gotten used to.
https://www.bleepingcomputer.com/news/microsoft/microsoft-ap...
> VSCode
Which is on top of 'Chrome'.
Interesting sandwich: Google-Microsoft-Google.
similar sandwich is Angular-Typescript-V8
Oh no, not another VSCode fork...
Looks like they managed to make it Mac only in the process.
I would love this to be true to further make fun of Electron, but I don't think it's Mac only.
https://antigravity.google/download
Ahh you’re right. The UX of that site is bad on my phone.
Many software engineering principles have been broken.
Google using Electron tells us that quality control is completely out of the window.
Unbelievable.
Native app dev is covered in red tape and puts you at the mercy of Apple etc. It's unfortunate that things are so inefficient now, but competition is good, and native platforms can get good.
> Native app dev is covered in red tape and puts you at the mercy of Apple etc.
... Aren't we talking about a programming IDE here? When did mobile become anything like the primary market for that? Are people expected to sit around for hours inputting symbols with an OSK?
Mac. It's a pain to target, and I don't mean cause of the code signing.
If Zed can make it, even with completely new stack, why not Google?
What are all this LLMs for when everything is just fork of electron app? Does not look like good marketing.
I tried writing a native Windows app using WinUI 3.
I wasted a day on trying to get some PNGs to render correctly, but no matter the config I used, the colors came out wrongly oversaturated.
I used Tauri with a WebView, and the app was rendering the images perfectly fine. On top of that the UI looked much better, and I was done in half the time I spend trying to fix the rendering issue in WinUI 3.
Never again will I go native.
To be fair WinUI3 is particularly bad. If you're forced to install a 100MB runtime anyway, might as well make that runtime be a browser.
Since Tauri uses the system's WebView, it shouldn't be that big. The docs say a minimal app could be 600 KB.
Making a VSCode fork is probably the wrong direction at this point in time. The future of agentic coding should need less support for code editor related functionality, and could eventually primarily support viewing code rather than editing code. There's a lot more flexibility in UI starting from scratch, and personally I want to see a UI that allows flexible manipulation of context and code changes with multiple agents.
GitHub is building a UI like this. I like it. I sometimes need the full IDE, but plenty of times don't. It's nice to be able to easily see what the agent is up to and converse with it in real-time while reviewing it's outputs.
"Congratulations, you have been elevated to manager to agents."
That's not exactly really where I hoped my career would lead. It's like managing junior developers, but without having nice people to work with.
"junior developers" is a convenient label, it is incorrect but it will take a bit until we come up something that describes entities that:
- can write code
- tireless
- have no aspirations
- have no stylistic or architectural preferences
- have massive, but at the same time not well defined, body of knowledge
- have no intrinsic memories of past interactions.
- change in unexpected ways when underlying models change
- ...
Edit: Drones? Drains?
- don't learn from what you tell them
- don't have career growth that you can feel good about having contributed to
- don't have a genuine interest in accomplishment or team goals
- have no past and no future. When you change companies, they won't recognize you in the hall.
- no ownership over results. If they make a mistake, they won't suffer.
Sounds like my teammates.
- don't learn from what you tell them
Whenever I have a model fix something new I ask it to update the markdown implementation guides I have in the docs folder in my projects. I add these files to context as needed. I have one for implementing routes and one for implementing backend tests and so on.
They then know how to do stuff in the future in my projects.
They still aren't learning. You're learning and then telling them to incorporate your learnings. They aren't able to remember this so you need to remind them each day.
That sounds a lot like '50 First Dates' but for programming.
> They aren't able to remember this
Yes, this is something people using LLMs for coding probably pick up on the first day. They're not "learning" as humans do obviously. Instead, the process is that you figure out what was missing from the first message you sent where they got something wrong, change it, and then restart from beginning. The "learning" is you keeping track of what you need to include in the context, how that process exactly works, is up to you. For some it's very automatic, and you don't add/remove things yourself, for others is keeping a text file around they copy-paste into a chat UI.
This is what people mean when they say "you can kind of do "learning" (not literally) for LLMs"
That is true, but does it actually matter if the outcome is the same? GP is saying they don't need to remind them.
Can this additional prompt from you also be automated? I do this too but I forget sometimes. I don't know if a general rule will be enough ?
> I add these files to context as needed.
Key words are these.
> They then know how to do stuff in the future in my projects.
No. No, they don't. Every new session is a blank slate, and you have to feed those markdown files manually to their context.
Imagine having these complaints about a screwdriver
It's a tool, not an intelligent being
Yeah, if my screwdriver undid the changes I just made to my mower, constantly ignored my desire to unscrew screws and instead punched a hole in my carb - I'd be throwing that screwdriver in the garbage.
I do not need to babysit my screwdriver.
Yet.
Next year there will be AI screwdriver your employer force you to use.
At first, I thought “ponector’s forgotten to add the /s”
Then I realised that this will actually happen, and was sadly reminded we’re now in the post-sarcasm era.
And you can buy AI screwdriver today from Amazon!
- don't learn from what you tell them
We'll fix that, eventually.
- don't have career growth that you can feel good about having contributed to
Humans are on the verge of building machines that are smarter than we are. I feel pretty goddamned awesome about that. It's what we're supposed to be doing.
- don't have a genuine interest in accomplishment or team goals
Easy to train for, if it turns out to be necessary. I'd always assumed that a competitive drive would be necessary in order to achieve or at least simulate human-level intelligence, but things don't seem to be playing out that way.
- have no past and no future. When you change companies, they won't recognize you in the hall.
Or on the picket line.
- no ownership over results. If they make a mistake, they won't suffer.
Good deal. Less human suffering is usually worth striving for.
Replace suffering with caring and have your AI write that again.
Why would I do a goofy thing like that?
> Humans are on the verge of building machines that are smarter than we are. I feel pretty goddamned awesome about that. It's what we're supposed to be doing.
It's also the premise of The Matrix. I feel pretty goddamned uneasy about that.
(Shrug) There are other sources of inspiration besides dystopic sci-fi movies. There's the Biblical story of the Tower of Babel, for instance. Better not work on language translation, which after all is how the whole LLM thing got started.
Sometimes fiction went in the wrong direction. Sometimes it didn't go far enough.
In any case, the matrix wasn't my inspiration here, but it is a pithy way to describe the concept. It's hard to imagine how humans maintain relevancy if we really do manage to invent something smarter than us. It could be that my imagination is limited though. I've been accused of that before.
> We'll fix that, eventually.
> Humans are on the verge of building machines that are smarter than we are.
You're not describing a system that exists. You're describing a system that might exist in some sci-fi fantasy future. You might as well be saying "there's no point learning to code because soon the rapture will come".
That particular future exists now, it's just not evenly distributed. Gemini 2.5 Pro Thinking is already as good at programming as I am. Architecture, probably not, but give it time. It's far better at math than I am, and at least as good at writing.
You'd also have no intrinsic memory of past interactions if we removed your hippocampus.
Coincidentally, the hippocampus looks like a seahorse (emoji). It's all connected.
> the hippocampus looks like a seahorse
Not to mention; hippocampus literally means "seahorse" in Greek. I knew neither of those things before today, thanks!
- constantly ignore your advice
- constantly give wrong answers, with surprising confidence
- constantly apologize, then make the same mistake again immediately
- constantly forget what you just told them
- ...
I describe them in the claude training I'm doing for my company as: super smart, infinitely patient, overeager interns
Sometimes smart sometimes the opposite, though. Perhaps due to memory loss.
Not sure "smart" or "dumb" are even the right axis to be judging them by, seems like intrinsically human traits.
Robots ?
Sounds like a junior developer?
They can usually write code, but not that well. They have lots of energy and little to say about architecture and style. Don't have a well defined body of knowledge and have no experience. Individual juniors don't change, but the cast members of your junior cohort regularly do.
"brooms"
*distant sound of the sorcerer's apprentice is heard*
I'm at one of those companies where we're forced to be in the office.
NO ONE TALKS TO EACH OTHER unless absolutely necessary for work.
We get on Zooms to talk. Even with the person 1 cubicle over.
> We get on Zooms to talk. Even with the person 1 cubicle over.
Who normalized this?!!
I actually do find there is a subset of meetings that are far more productive on Zoom. We can be voice chatting on one screen, share another screen and both be able to type, record notes, pull up side research without interrupting the conversation. It's a bit closer to co-working than a meeting but it hits a sweetspot for me.
I know it happens when you're working with people who may be at home or another location.
> We get on Zooms to talk. Even with the person 1 cubicle over.
But why? Required? Culture? Maybe it's the company?
Why?
I thought you were joking until I saw the video where this is an actual quote.
> It's like managing junior developers, but without having nice people to work with.
Nice? I thought all sycophant LLMs were exceedingly nice.
For me, it feels so fake that I'd rather have it not try to be nice. I guess I've gotten a bit used to it and ignore it for most parts.
I used to be really excited about "agents" when I thought people were trying to build actual agents like we've been working on in the CS field for decades now.
It's clear now that "agents" in the context of "AI" is really about answering the question "How can we make users make 10x more calls to our models in a way that makes it feel like we're not just squeezing money out of them?" I've seen so many people that think setting some "agents" of on a minutes to hours long task of basically just driving up internal KPIs at LLM providers is cutting edge work.
The problem is, I haven't seen any evidence at all that spending 10x the number of API calls on an agent results in anything closer to useful than last year when people where purely vibe coding all the time. At least then people would interactively learn about the slop they were building.
It's astounding to watch a coworker walk though through a PR with hundreds of added new files and repeatedly mention "I'm not sure if these actually work, but it does look like there's something here".
Now I'm sure I'll get some fantastic "no true Scotsman" replies about how my coworkers must not be skilled enough or how they need to follow xyz pattern, but the entire point of AI was to remove the need for specialize skills and make everyone 10x more productive.
Not to mention that the shift in focus on "agents" is also useful in detracting from clearly diminishing returns on foundation models. I just hope there are enough people that still remember how to code (and think in some cases) to rebuild when this house of cards falls apart.
> but the entire point of AI was to remove the need for specialize skills and make everyone 10x more productive.
At least for programming tools, for everything (well, the vast majority, at least) that is sold that way—since long before generative AI—it actually succeeds or fails based not on whether it eliminates need for specialized skills and makes everyone more productive, but whether it further rewards specialized skills, and makes the people who devote time to learning it more productive than if they devoted the same time to learning something else.
OR - think about the junior developers being managers of agents, and you are still a manager of junior developers. This is not zero sum!
I am essentially in this exact role. The junior developers simply don't have the experience to evaluate the output of the agents. You wind up with a lot of slop in PRs. People can't justify why they did something. I've seen whole PRs closed, work redone, and opened anew because they were 70% garbage. Every other comment was asking "why is this here? it has nothing to do with the ticket."
Sadly, this is not sustainable and I am not sure what I'm going to do.
I enjoy getting into a good flow state and pounding out clever and elegant code but watching a good LLM generate code according to my specs and refining it is also enjoyable. I've been burning through $250 of free Claude Code Web credits and having multiple workers running at the same time is fun.
Also, agents have no capacity to learn.
They have a capacity to "learn", it's just WAY MORE INVOLVED than how humans learn.
With a human, you give them feedback or advice and generally by the 2nd or 3rd time the same kind of thing happens they can figure it out and improve. With an LLM, you have to specifically setup a convoluted (and potentially financially and electrical power expensive) system in order to provide MANY MORE examples of how to improve via fine tuning or other training actions.
Depending on your definition of "learn", you can also use something akin to ChatGPT's Memory feature. When you teach it something, just have it take notes on how to do that thing and include its notes in the system prompt for next time. Much cheaper than fine-tuning. But still obviously far less efficient and effective than human learning.
I think it’s reasonable to say that different approaches to learning is some kind of spectrum, but that contemporary fine tuning isn’t on that spectrum at all.
> With an LLM, you have to specifically setup a convoluted (and potentially financially and electrical power expensive) system in order to provide MANY MORE examples of how to improve via fine tuning or other training actions.
The only way that an AI model can "learn" is during model creation, which is then fixed. Any "instructions" or other data or "correcting" you give the model is just part of the context window.
Hold that thought.
As an introvert, that's a pro, not a con.
The burden of human interaction is removed from building.
I'm an introvert and I love working with nice people.
I just need some time by myself to recharge after all the social interactions.
As a team lead, working with people is so... cumbersome. They need time to recharge, lots of encouragement, and a nice place to work in. Give me a coding agent any time!
They don't like working on the cumbersome tickets, writing tests, documentation. Talking to businesspeople.
As a PM working with team leads is so cumbersome...
Being antisocial isn't an introvert thing. I'm incredibly introverted and still love having time interacting with people
2020: every day a new JS framework is announced
2024: every day a new Chrome fork browser is announced
2025: every day a new AI IDE vscode fork is announced
2020: every day a new electron fork is announced
2024: every day a new electron fork is announced
2025: every day a new electron fork is announced
There's only one electron. It looks like separate ones because it's traveling backwards and forwards in time.
>vscode fork
I wonder why they are not trying to fixup something based on their own GUI stacks like Flutter or Compose Multiplatform.
It seems only Zed is truly innovating in this space.
Well it's a helluva lot faster to make for one. For two, just about everyone knows how to navigate in vscode by now. Reducing the barrier of entry has obvious advantages.
I have Cursor at work. The only element of the interface I'm using (and know how to use) is the chat window.
IMO, it's an absolutely crappy IDE, crappy editor, with absolutely incomprehensible hostile UI.
I have almost two decades of experience with Vim, Emacs and IntelliJ. FWIW, I was able to easily find my ways in helix, kakoune and Zed.
https://cursor.com/cli
Really? Can you say what you hate it about it pls
I'd like to retract my critique.
I just opened the app to see what else I can bring up, and while clicking through UI I noticed I had some crappy key bindings extension installed, which apparently caused many of my annoyances.
I've probably installed it very long ago, or even by accident.
For example, I was always annoyed that open file/directory shortcut (one of most common operations) is not assigned and requires mouse interaction -- fixed by disabling the extension. Go to file shortcuts does something completely different -- fixed by disabling the extensions.
I likely won't adopt Cursor as my main IDE/Editor, but it's miles better than I thought just an hour ago.
Thanks for your question :D
Not the person you asked, but I hate how it screws up keyboard shortcuts. It overrode the delete line shortcut with its own inline chat one, for example.
Decided to ditch it for claude code right after that, since I cannot be bothered to go over the entire list of keyboard shortcuts and see what else it overrode/broke.
I've found that annoying too, but you can always rebind them as you wish. It's only a few new keybinds that get in the way of my muscle memory.
That said I also have moved to CLI agents like Claude Code and Codex because I just find them more convenient and, for whatever reason, more intelligent and more likely to correctly do what I request.
From the top of my head in no specific order
- Icons on the toolbar in the left panel have no labels or even tooltips. No way to know what they do without clicking and checking.
- Space in the file explorer in the left panel opens a file (haven't noticed such behavior in other editors -- totally unexpected).
- Maybe that's the artifact of me installing Vim plugin, but Keyboard shortcuts displayed in the main menu don't do what they say they do.
- It often offers installing some plugins, and I've absolutely no idea why, and what will happen if I do or if I don't.
I'm talking about Cursor, which I assume is exactly like VS Code. Tried VS Code only once very long ago.
I think the left toolbar icons do have tooltips, but they show up pretty slowly for me (2-3 seconds).
These new editors are trying to differentiate themselves via their AI features. Working on the core editor may waste resources that could have been better spent improving the AI features.
Until someone finally figures out that we need to rethink editors from the ground up to support different sort of operations and editing experience, to better facilitate LLMs doing work as agents.
But we're probably 1-2 years away from there still, so we'll live with skinned-forks, VSCode extensions and TUIs for now.
How would you say Zed is innovating? Never heard of it, just taking a peek now.
Zed team is writing their own in-house GUI stack [1] that leverages the computer's GPU with minimal middleware in-between. It's a lot of work short-term but IMO the payoff would be huge if they establish themselves. I imagine they could poke into the user-facing OS sector if their human-agent interaction is smooth. (I have not tried it yet though)
[1]: https://www.gpui.rs
What about the text editing part?
Sublime did that almost 20 years ago.
It's written from scratch in Rust. It's super fast, polished, etc. A world of difference compared to VSCode.
it's fast
I am very sensitive to input latency and performance but after comparing Zed and VS Code for a while I really couldn't find any reason to stick with Zed. It's been a year or so since I last tried it but VSC just lets me do way more while still, IMO, having a nice, clean UI. I never notice any performance or key input latency with VSC.
If you work in a team there are some pretty good remote collaboration tooling built into Zed.
Making an IDE sounds like an insane amount of effort.
FWIW, the Fuchsia team was working on an editor that had a Flutter UI when run in Fuchsia:
https://xi-editor.io/frontends.html
> I wonder why they are not trying to fixup something extremely complex that only a handful players managed to get right using gui stacks made with only mobile in mind that are desperately trying to catch up to desktop now
We need a VS Code fork that just exposes more interfaces, and does nothing else. Then all these forks could just use that with power extensions, and it'd force Microsoft to change its behavior.
You just described Eclipse RCP.
The issue with Eclipse and that approach is the complexity of mixing plugins to do everything, which kills the UX.
When VSCode started, the differentiator from Atom and Eclipse was that the extension points were intentionally limited to optimize the user experience. But with the introduction of Copilot that wasn’t enough, hence the amount of forks.
I think that the Zed approach of having a common protocol to talk with agents (like a LSP but for agents) is much better. The only thing that holds me from switching to Zed is that so far my experience using it hasn’t been that good (it still has many rough edges)
There's actually an ACP that has some traction as well, though I don't think it solves the problem of what the agent can do in vscode: https://agentcommunicationprotocol.dev/introduction/welcome
there are now 15 competing standards
Why is it so hard for these to be VSCode extensions and not forks?
Microsoft has very specific constraints on what extensions can and can't do, it's not a free for all. They're actively defending their mote by allowing Copilot to do things in a way that extensions couldn't. That's why all the serious contenders make a fork, it's simply not possible to have the same integration otherwise.
A bit tangential, but I asked the new Gemini model about this, and it immediately traced back this quote to your username: https://gemini.google.com/share/144b46094d6e
Creepy stuff :)
Still it would be a lot wiser if all the forkers would do one 'AI-enabled' fork together that exposes all the extras that copilot gets. The barrier for testing would be much lower and all the extension makers would also jump onto the train. Likely MS would finally give in and make all the extras available for everyone. But all the fragmentation only helps MS.
What exactly are the "extras" that Copilot gets?
I've had a Github Copilot subscription from work for 1yr+ and switch between the official Copilot and Roo/Kilo Code from time to time. The official Copilot extension has improved a lot in the last 3-6 months but I can't recall ever seeing Copilot do something that Roo/Kilo can't do, or am I missing something obvious?
They're all going to have quite divergent opinions on how to structure the fork and various other design decisions that would inevitably lead to forks of the fork again.
I think forking VS Code is probably the most sensible strategy and I think that will remain the case for many years. Really, I don't think it's changing until AI agents get so ridiculously good that you can vibe code an entire full-featured polished editor in one or a few sessions with an LLM. Then we'll be seeing lots of de novo editors.
This isn't true. I've gotten Claude and its forks to do things the same way as Copilot.
Which is a bit odd given that Claude code extension in VSCode is by far the best agent integration into a codebase that I know of.
The Claude Code extension on VS Code does very little (too little in my opinion). The integration level with agentic functionality provided by Antigravity goes much deeper in my 20 minutes or so of playing with it. The biggest value pieces I see is: Agent Manager window which provides a unified view of all my agents running across all my workspaces (!) where I can quickly approve or respond to followup questions and quickly brings me to the code in context for each agent, additionally, I can select a piece of code and comment on it inline and that comment gets sent to the correct, active agent. These two things alone are items which I have been looking for... Too bad I only have approval to use Claude Code at work. This looks promising.
That’s all very silly and unnecessary in my opinion. Agentic change by change diffs are optimal for a professional.
I don't even know what the Claude Code extension does in vscode. I have it installed but hell if I know what it's doing. I run Claude in one of vscode's terminals, and do everything through there. I do see (sometimes) diffs pop up in the IDE, I guess that's the extent of this integration.
*moat
AIUI the forks are required because Microsoft is gatekeeping functionality used by Copilot from extensions so they can't be used by these agents.
> AIUI the forks are required because Microsoft is gatekeeping functionality used by Copilot from extensions so they can't be used by these agents. reply
I always wonder how this works legally. VSCode needs to comply with the LGPL (it's based on Chromium/Blink which is LGPL) ; they should provide the entire sources that allow us to rebuild our own "official" VSCode binary
I think that MS provides these extensions as plugins. The core of VSCode is OSS, it's the extra bits that have a different license.
Could you give an example of what they're gatekeeping for Copilot exclusively? I'm kinda confused because Copilot in VS Code isn't exactly a powerhouse of unique features in my experience, it still feels well behind Roo/Kilo Code in most ways I can think of, although much closer to the competition than it was a year ago.
> AI UI
2026: every day ...
I was going to ask why all these companies choose to fork the entire IDE rather than just writing an extension like every other sane developer, and this response is the most believable reason why.
But Microsoft made VSCode lol, I think being able to gatekeep things like that shouldn’t allow a billion dollar company just reuse all of your code instead of making their own IDE
Most of VSCode (yes, most) is a mishmash of other OSS products, including Google Chromium! By that logic VSCode itself shouldn't exist.
> 2024: every day a new Chrome fork browser is announced
I think this was more accurate around 2012. My local tech magazine had their own fork and they attached CD with the magazine which included the browser.
> 2024: every day a new Chrome fork browser is announced
This is still happening. Didn't you see OpenAI's release of Atlas?
This whole blog post is seemingly about Google, not about the user. "Why We Built Antigravity" etc. "We want Antigravity to be the home base for software development in the era of agents" - cool, why would I as the user care about that?
You wouldn’t. It’s made to suck out investor money and show that google does something, not to actually bring value.
My crystal ball says it will be shutdown next year.
Most of the AI products are not for the end user, they are just signaling shareholders and possible investors that the company is on the hype.
This kind of cynicism is wild to me. Of course most AI products (and products in general) are for end users. Especially for a company like Google--they need to do everything they can to win the AI wars, and that means winning adoption for their AI models.
http://killedbygoogle.com/ - most Google products are for the temporary career advancement of some exec or product lead.
Their only real product is advertising, everything else is a pretense to capture users attention and behaviors that they can auction off.
There is also mechanism inside Google that rewards teams that launches new products, more than the teams that actually maintain existing ones.
Google is still looking for investors?
Of course, Alphabet exists to give returns to their shareholders.
I know Google is quick to shut things down but these ultra-cynical ultra-skeptical HN takes are so tiresome at this point.
This is what Google is like though. It is practically part of their corporate DNA.
I do not believe that Google Antigravity is aimed at wooing investors. I believe it is intended to be a genuine superior alternative to Cursor and Kiro etc. and is attempting to provide the best AI coding experience for the average developer.
Most of the other people (so far) in this sub-thread do not think this. They essentially have a conspiratorial view on it.
funny my magic 8 ball says the same thing!
After using Google AI studio, Google Vertex, and Google Gemini Chat I honestly can't wait to use Google Antigravity!
edit: Also Jules...
snark off:
I think the Google PMs should have coffee together and see if all of this sprawl makes any sense.
It does?
Google AI studio is their developer dashboard.
Google Vertex is their equivalent of Amazon Bedrock.
Google Gemini Chat is their ChatGPT app for normies.
Google Antigravity is their Cursor equivalent.
I agree what you’ve listed makes sense as a product portfolio.
But AI Studio is getting vibe coding tools. AI Studio also has a API that competes with Vertex. They have IDE plugins for existing IDEs to expose Chat, Agents, etc. They also have Gemini CLI for when those don’t work. There is also Firebase Studio, a browser based IDE for vibe coding. Jules, a browser based code agent orchestration tool. Opal, a node-based tool to build AI… things? Stich, a tool to build UIs. Colab with AI for a different type of coding. Notebook LM for AI research (many features now available in Gemini App). AI Overviews and AI mode in search, which now feature a generic chat interface.
Thats just new stuff, and not including all the existing products (Gmail, Home) that have Gemini added.
This is the benefit of a big company vs startups. They can build out a product for every type of user and every user journey, at once.
Don't forget Gemini CLI
In another 2 years we'll probably be back to just "Google" as digital agent that can do any research, creative, or coding task you can imagine.
I concede.
And In practice, when I needed to use one of their models for a small project, it turned out that the only sane way is to go via OpenRouter…
> Google Vertex is their equivalent of Amazon Bedrock
Well, that clears that up.
In "real world" you don't use OpenAI or Anthropic API directly—you are forced to use AWS, GCP, or Azure. Each of these has its own service for running LLMs, which is conceptually the same as using OpenAI or Anthropic API directly, but with much worse DX. For AWS it's called Bedrock, for GCP—Vertex, and for Azure it's AI Foundry I believe. They also may offer complementary features like prompt management, evals, etc, but from what I've seen so far it's all crap.
Also gemini-cli (terrible)
Google ADK (agent development kit, awesome)
The also launched a coding agent Jules: https://jules.google/
Jules is the first and only one to add a full API, which I've found very beneficial. It lets you integrate agentic coding features into web apps quite nicely. (In theory you could always hack your own thing together with Claude Code or Codex to achieve a similar effect but a cloud agent with an API saves a lot of effort.)
Google ADK is real nice and gives you an API as well (also web browser and terminal prompt)
Jules is nifty. Weirdly heavy on the browser CPU.
you forgot jules
Great point!
Remember took my a while early in my career from changing my resume away from saying "I want to do this at my next job and make a lot of money" and towards "here is how I can make money and save costs for your company".
Google didn't learn that lesson here. They are describing why us using Antigravity is good for Google, not why us using Antigravity is good for us.
More accurately, it should be neither about Google nor about the user, but about the product. Describe what the product is and does, don’t make assumptions about the user, and let the user be the judge of it.
I swear most of these pages is to sell this to companies so they can force it onto developers.
The whole webpage looks like something from Apple.
All these companies were built by self-referential narcissists, and it seems to be their culture at the core.
On the pricing page it says that for public preview they are offering a free individual plan with "generous rate limits". I gave it an HTML file and asked it to create Jinja templates from it and 2 minutes later (still planning, no additional prompt) I got this:
> Model quota limit exceeded. You have reached the quota limit for this model.
I think the models are under high load right now, and not working properly.
> html
Would be willing to bet this is the issue. Adding html files to context for gemini models results in a ton of token use.
why?
EDIT: why must users care?
Gotta learn all the quirks of the model before it's replaced in 8 minutes.
Quirks? like context window?
I'm saying it's egregious to expect all users to know the fact that an HTML document, for some reason, uses an enormous amount of context in an LLM designed specifically for working with code.
https://stackoverflow.com/questions/1732348/regex-match-open...
The accepted answer is one that doesn’t care about the questioner‘s use case and instead gives a pretty excessive "Don‘t do it"
It does also give the right solution, using an xml parser.
We don’t know the use case.
Maybe the questioner is also in full control of the HTML creation and they don’t need a parser for all possible HTML edge cases.
Jevons paradox - https://en.wikipedia.org/wiki/Jevons_paradox
It's the same problem with OpenRouter's free tiers for a long time. If something is truly $0 and widely available, people will absolutely bleed it dry.
Same here. I tried to build a super simple iOS App in antigravity and I was out of quota before it finished. The whole thing was a couple of files and a few hundred lines of code.
That'd be my experience with gemini cli.
I had pretty much the same experience.
The agentic spam is exhausting. I just wanted to code.
Too early in my career to not give a shit and retire, but too late be excited about these things and eager to learn. What a time...
Same here - completely relate.
One thing I’ve noticed though that actually coding (without the use of AI; maybe a bit of tab auto-complete) is that I’m actually way faster when working in my domain than I am when using AI tools.
Everytime I use AI tools in my domain-expertise area, I find it ends up slowing me down. Introducing subtle bugs, me having to provide insane amount of context and details (at which point it becomes way faster to do it myself)
Just code and chill man - having spent the last 6 months really trying everything (all these context engineering strategies, agents, CLAUDE.md files on every directory, et, etc). It really easy still more productive to just code yourself if you know what you’re doing.
The thing I love most though - is having discussions with an LLM about an implementation, having it write some quick unit tests and performance tests for certain base cases, having it write a quick shell script, etc. things like this, it’s Amazing and makes me really enjoy programming since I save time and can focus on doing the actual fun stuff
it is absolutely poor skill, or disengenuous at best, for any coder to claim AI tools slow them down lol.
When I'm doing the coding myself, I'm at least making steady progress and the process is predictable. With LLMs, it's a crapshoot. I have to get the AI to understand what I want and may have try again multiple times, many times never succeeding, and end up writing a lot of text anyway. And in between, I'll have to read a lot of code that probably ends up being thrown away or heavily modified. This probably depends a lot on what kind of project one is working on, though.
But it's like you said, I like using LLMs for completing smaller parts or asking specific kind of help or having conversations about solutions, but for anything larger, it just feels like banging my head to a wall.
One of the core principles of my workflow (inspired by REPL development and some unix tools) is to start with a single file (for a function or the whole project). The I refactor the code to have a better organization and to improve reliability, especially as I'm handling more scenarios (and failure modes).
LLMs are not useful in this workflow, because they are too verbose. Their answers are generic and handle scenarios you don't even support yet. What's useful is good documentation (as in truthful) and the code if it's open.
This approach has worked really well in my career. It gives me KISS and YAGNI for free. And every line of code is purposeful and have a reason to be there.
Devs are starting to realize that the sweet spot for AI support in coding is on a small scale, i.e. extended code completion. Generating huge chunks of code is often not reliable enough except for some niches (such as simple greenfield projects which profit from millions of examples online).
For those discussions with the LLM do you just use Gemini chat or chat GPT etc... i.e. the chat interface?
You might as well be saying you want to make physical architecture draft drawings.
I feel this. Tbh I was excited about most of the previous tech fads I have come across. Sure, people got overzealous with them and also failed to realize that what F*NG needs is not what your three person startup needs, but there were always some good and interesting ideas there.
This just feels... a little too dystopian. Companies hoovered up the entirety more or less of all of our collective thoughts and writings and output and now want to sell it back to us- and I fear that cost is going to be extremely steep.
It's impressive, but at the same time, just feels like its going to somehow be a net detractor to society, and yet I feel I need to keep up with each new iteration or potentially get washed over and left behind by the wave.
I am somewhat fortunate to be towards the top of the pyramid and also in a position where I could theoretically ride off into the sunset, but I fear the societal implications and the pain that is going to come for vast numbers of people.
It's never too early to retire, there are plenty of other things to work on outside of tech
The robots have that covered
> The agentic spam is exhausting. I just wanted to code.
Ditto. And we still can.
I've yet to use an "agent", and still use a chat UI to an LLM in Emacs. I rely on these tools for design discussion, rough prototyping, and quick reference, but they still waste my time roughly a quarter of the time I use them. They have gotten better in the last year, though, and I've been able to broaden my reach into stacks and codebases I wouldn't have felt comfortable with before, which is good.
I just have no interest in "agents". I don't want to give these companies more access to my system and data, and I want to review every thing these tools generate. If this makes me slower than a vibe coder, that's intentional. Thankfully, there are still sane people and companies willing to pay me for this type of work, so I'm not worried about being displaced any time soon. Once that happens, I'll probably close up shop, figure out an alternative income stream, and continue coding as a hobby.
> Spin up agents to tackle routine tasks that take you out of your flow, such as codebase research, bug fixes, and backlog tasks.
The software of the future, where nobody on staff knows how anything is built, no one understands why anything breaks, and cruft multiplies exponentially.
But at least we're not taken out of our flow!
After a bunch of people leave the company it's already like nobody knows how anything is built. This seems like a good thing to accelerate understanding a codebase.
it's funny - nervous funny, not haha funny - that you think drawing a real issue like this out into the open would focus an organization on solving it.
Agents pretty good at describing how a particular feature works. It's not as dire as you make it seem.
> The software of the future,
:chuckles nervously:
If you're building something new you'll need some skilled people around
You can ask agents to identify and remove cruft. You can ask an agent why something is breaking -- to hypothesize potential causes and test them for validity. If you don't understand how something is built, you can ask the agent to give you an overview of the architecture and then dive into whatever part you want to explore more.
And it's not like any of your criticisms don't apply to human teams. They also let cruft develop, are confused by breakages, and don't understand the code because everyone on the original team has since left for another company.
> you can ask the agent to give you an overview of the architecture and then dive into whatever part you want to explore more.
This is actually a cool use that's being explored more and more. I first saw it in the wiki thing from the devin people, and now google released one as well.
Humans are just better at communicating about their process. They will spend hours talking over architectural decisions, implementation issues, writing technical details in commit messages and issue notes, and in this way they not only debug their decisions but socialize knowledge of both the code and the reasons it came to be that way. Communication and collaboration are the real adaptive skills of our species. To the extent AI can aid in those, it will be useful. To the extent it goes off and does everything in a silo, it will ultimately be ignored - much like many developers who attempt this.
I do think the primary strengths of genai are more in comprehension and troubleshooting than generating code - so far. These activities play into the collaboration and communication narrative. I would not trust an AI to clean up cruft or refactor a codebase unsupervised. Even if it did an excellent job, who would really know?
> Humans are just better at communicating about their process.
I wish that were true.
In my experience, most of the time they're not doing the things you talk about -- major architectural decisions don't get documented anywhere, commit messages give no "why", and the people who the knowledge got socialized to in unrecorded conversations then left the company.
If anything, LLM's seem to be far more consistent in documenting the rationales for design decisions, leaving clear comments in code and commit messages, etc. if you ask them to.
Unfortunately, humans generally are not better at communicating about their process, in my experience. Most engineers I know enjoy writing code, and hate documenting what they're doing. Git and issue-tracking have helped somewhat, but it's still very often about the "what" and not the "why this way".
"major architectural decisions don't get documented anywhere" "commit messages give no "why""
This is so far outside of common industry practices that I don't think your sentiment generalizes. Or perhaps your expectation of what should go in a single commit message is different from the rest of us...
LLMs, especially those with reasoning chains, are notoriously bad at explaining their thought process. This isn't vibes, it is empiricism: https://arxiv.org/abs/2305.04388
If you are genuinely working somewhere where the people around you are worse than LLMs at explaining and documenting their thought process, I would looking elsewhere. Can't imagine that is good for one's own development (or sanity).
I've worked everywhere from small startups to megacorps. The megacorps certainly do better with things like initial design documents that startups often skip entirely, but even then they're often largely out-of-date because nobody updates them. I can guarantee you that I am talking about common industry practices in consumer-facing apps.
I'm not really interested in what some academic paper has to say -- I use LLM's daily and see first-hand the quality of the documentation and explanations they produce.
I don't think there's any question that, as a general rule, LLM's do a much better job documenting what they're doing, and making it easy for people to read their code, with copious comments explaining what the code is doing and why. Engineers, on the other hand, have lots of competing priorities -- even when they want to document more, the thing needs to be shipped yesterday.
Alright, I'm glad to hear you've had a successful and rich professional career. We definitely agree that engineers generally fail to document when they have competing priorities, and that LLMs can be of use to help offload some of that work successfully.
Your initial comment made it sound like you were commenting on a genuine apples-for-apples comparisons between humans and LLMs, in a controlled setting. That's the place for empiricism, and I think dismissing studies examining such situations is a mistake.
A good warning flag for why that is a mistake is the recent article that showed engineers estimated LLMs sped them up by like 24%, but when measured they were actually slower by 17%. One should always examine whether or not the specifics of the study really applies to them--there is no "end all be all" in empiricism--but when in doubt the scientific method is our primary tool for determining what is actually going on.
But we can just vibe it lol. Fwiw, the parent comment's claims line up more with my experience than yours. Leave an agent running for "hours" (as specified in the comment) coming up with architectural choices, ask it to document all of it, and then come back and see it is a massive mess. I have yet to have a colleague do that, without reaching out and saying "help I'm out of my depth".
Sure, you can ask the agents to "identify and remove cruft" but I never have any confidence that they actually do that reliably. Sometimes it works. Mostly they just burn tokens, in my experience.
> And it's not like any of your criticisms don't apply to human teams.
Every time the limitations of AI are discussed, we see this unfair standard applied: ideal AI output is compared to the worst human output. We get it, people suck, and sometimes the AI is better.
At least the ways that humans screw up are predictable to me. And I rarely find myself in a gaslighting session with my coworkers where I repeatedly have to tell them that they're doing it wrong, only to be met with "oh my, you're so right!" and watch them re-write the same flawed code over and over again.
Because what we need is not lazy people - we need lazy people with AI? How is this even a justification?
Sorry, where did "lazy people" come from? Nobody's talking about anybody being lazy.
I just like that evolution doesnt really care. People can opine on laziness and proper methodology. Its handwaving compared to how things shake out.
Nature does select for laziness. The laziest state that can outpace entropy in diverse ways? Ideal selection.
its so funny watching crusty devs malding over AI. do you all write assembly code for your apps too? because who can trust those magic compiler things right?
AI is just another abstraction that allows you to do 10x with 1x the effort. either you'll learn to master it or you'll become obsolete
How is a compiler and an LLM equivalent abstractions? I'm also seriously doubtful of the 10x claim any time someone brings it up when AI is being discussed. I'm sure they can be 10x for some problems but they can also be -10x. They're not as consistently predictable (and good) like compilers are.
The "learn to master it or become obsolete" sentiment also doesn't make a lot of sense to me. Isn't the whole point of AI as a technology that people shouldn't need to spend years mastering a craft to do something well? It's literally trying to automate intelligence.
This argument warrants introspection for "crusty devs", but also has holes. A compiler is tightly engineered and dependable. I have never had to write assembly because I know that my compiled code 100% represents my abstract code and any functional problems are in my abstract code. That is not true in AI coding. Additionally, AI coding is not just an abstraction over code, but an abstraction over understanding. When my code compiles, I don't need to worry that the compiler misunderstood my intention.
I'm not saying AI is not a useful abstraction, but I am saying that it is not a trustworthy one.
I’d worry about mastering the shift key, first.
The software of the future, where nobody on staff knows how anything is built
Doesn't this apply to people who code in high level languages?
The increasing levels of abstraction work only as long as the abstractions are deterministic (with some limited exceptions - i.e. branch prediction/preloading at CPU level, etc). You can still get into issues with leaky abstractions, but generally they are quite rare in established high->low level language transformations.
This is more akin to manager-level view of the code (who need developers to go and look at the "deterministic" instructions); the abstraction is a lot lot more leaky than high->low level languages.
In the 00s I saw so many C codebases with hand-rolled linked lists where dynamically resized arrays would be more appropriate, "should be big enough" static allocations with no real idea of how to determine that size, etc. Hardly anyone seemed to have a practical understanding of hashes. When you use a higher level language, you get steered towards the practical, fundamental data structures more or less automatically.
even JS doesn't churn as fast as the models powering vibe coding, and that cut & paste node app is still deterministic, compared to what happens when the next version of the model looks at AI-generated code from two years ago...
> Neither engenders user trust in the work that the agent undertook. Antigravity provides context on agentic work at a more natural task-level abstraction, with the necessary and sufficient set of artifacts and verification results, for the user to gain that trust.
I'm going to need an AI summary of this page to even start comprehending this... It doesn't help that the scrolling makes me nauseous, just like real anti-gravity probably would.
Then there's this:
"A more intuitive task-based approach to monitoring agent activity, presenting you with essential artifacts and verification results to build trust."
The whole thing around "trust" is really weird. Why would I as a user care about that? It's not as if LLMs are perfect oracles and the only thing between us and a brave new world is blind trust in whatever the LLM outputs.
I'd say that because right now, verifying the LLM output (meaning: not trusting it) is a huge bottleneck (rightfully so). I guess they are trying to convince people that this bottleneck is no more, with this IDE.
Written by AI now summarise and explained by AI.
wow, you weren't kidding about the scroll induced nausea.
Translation: "don't trouble your little brain actually trying to read the code that our model produced. we can show you pictures of the output instead."
I'll be honest - this doesn't look half-bad.
It really seems like it's just standardizing into a first-class UI what a lot of people have already been doing.
I don't think I'm the target for this - I already use Claude Code with jj workspaces and a mostly design-doc first workflow, and I don't see why I would switch to this, but I think this could be quite useful for people who don't want to dive in so deep and combine raw tooling themselves.
Hadn't heard of jj, it's a source control tool that advertises that it's fast and compatible with the git on disk format. https://github.com/jj-vcs/jj
As a daily Claude Code and Codex user, I've really got to start getting into jj. I keep telling myself I will but I'm just so used to git.
Can you elaborate on how you personally use jj workspaces with command-line coding agents?
Sure. Honestly I think you can get the same with git work trees, though I haven't tried.
After a couple iterations on this, I've ended up having claude code vibe-code a helper CLI in Go for me which I can invoke with `ontheside <new-workspace-name> <base-change>` and will
- create a new jj workspace based on the given change
- create a docker container configured with everything my unit tests need to run, my claude code config mounted, and jj configured
- it also sets up a claude code hook to run `jj` (no arguments) every time it changes a file, so that jj does a snapshot
- finishes by starting an interactive claude code session with `--dangerously-skip-permissions`
- it also cleans it all up once I exit the claude code session and fish shell that's running it
With this I can have Claude Code working asynchronously on something, while I can review (or peek) the changes in batch from my main editor by running `jj show <change-id>` / `jj diff -r "..."` (which in my case opens it up in the Goland multi-file diff viewer). I can also easily switch to the change it's working on in my main editor, to make some manual modifications if necessary.
This is, in general, primarily for "in the background async stuff" I want to have it work on. Most of the time I just have a dead-normal claude code session running in my main workspace.
Minor self-plug - if you want, I posted a jj intro article a while ago[0], though it doesn't include my current workspace usage.
[0]: https://kubamartin.com/posts/introduction-to-the-jujutsu-vcs...
The Agent Manager view providing a unified view of all active agents and allowing you to immediately respond to any approval requests or followup questions looks very useful regardless of which VCS you're using under the covers. Am I missing something here that jj does?
See my setup detailed in another sibling comment of yours, jj is just a small part of it, and you can probably get that with git too.
I’m already at full mental capacity planning and reviewing the work of two agents (one foreground which almost never asks for approval, and one background which never asks for approval).
I don’t really need the ability to juggle more of them, and noticing their messages is not a bottleneck for me, while I’m happy with the customizability and adaptability of my raw’er workflow.
Maybe if they’re as slow as codex…
Fair enough. I use git worktrees (with a script that creates the git branch, worktree and opens a new vs code workspace). You're right, managing more than about two active sessions at once is probably the limit though I'm somewhat hopeful that better tooling similar to the Agent Manager window here would allow me to scale a bit past that especially if some of those sessions are more design explorations.
It's a good start, but I found it lacking compared to VSCode+CLINE+Gemini Pro.
- A new "AI" IDE announced
- It's VS Code
Like clockwork!
This is why I have much respect for the Zed team as they are chasing originality, not just slap something onto VS Code and call it a new IDE.
If both work equally well then why does it matter whether Zed or VSCode is under the hood?
I love Zed code. Sad that Sublime Text is not keeping up.
Once the UI soup around AI dev use has settled (and it's getting closer) I bet you we will see native apps with c/c++/zig/rust backends that render so much faster on all the junctions that aren't roundtrip limited (and yes, that will still matter to many people).
These are two of the fastest editors, and I use both, but I don't think they're for the same thing: Zed is for multi-file projects with moderate IDE abstractions (Worse than Jetbrains; better than most others); Sublime is for editing one-off files with syntax highlighting.
On the contrary, I am glad that the Sublime team did not fall for the trends and avoided adding AI slop into the editor
> VSCode
You mean Chromium wrapper?
Chromium? You mean OS wrapper?
Is the extension system in VSCode not powerful enough to make these just normal extensions for a vanilla VSCode executable? Or is everyone just going for lock in, since if you download MyFork, you can't start using some other extension that uses OtherGuysModel?
Back when Cursor was new (before literally everything was AI hype) they explicitly called out that they wanted to do more in-depth integration with the editor than was possible with just the extension APIs.
Presumably that hasn't changed much. If you want to do any large-scale edits of the UI you need to spin up a fork.
Correct. One could say the same thing for browsers. I suppose on the one hand it's good that it's relatively easy to spin up a new project like this, on the other hand one must swear allegiance to their large software company of choice.
It's also an absolutely basic fork. They haven't even bothered with a custom theme, or custom UI, it's just vscode with an agents window slapped on top.
Weirdly, out of all the vscode forks the best UI is probably bytedance's TRAE
How many forks of VS Code am I supposed to have installed at this point?
You'll need a Chromium based app to count the installs for you.
My YC2026 startups are an AI agent that automatically manages your VSCode forks, and a public safety computer vision app for smart glasses that predicts whether someone can afford a lawyer based on their skin color
They'll have to compete with these other pending funded companies:
- ai therapist for your ai agents
- genetic ai agent container orchestration; only the best results survive so they have to fight for their virtual lives
- prompt engineering as a service
- social media post generator so you can post "what if ai fundamentally changes everything about how people interact with software" think pieces even faster
- vector db but actually json for some reason
(Edit: formatting)
> genetic ai agent container orchestration; only the best results survive so they have to fight for their virtual lives
AI Agent Orchestration Battle Bots. Old school VMs are the brute tanks just slowly ramming everybody off the field. A swarm of erratically behaving lightweight K8s Pods madly slicing and dicing everything coming on their way. Winner takes control of the host capacity.
I might need this in my life.
Don't forget there's 2 or 3 of each in YC2026.
yes
With all due to respect to the folks working on Antigravity, this feels like a vibe-coded VSCode fork to me. Font sizes, icon sizes, panel sizes are all over the place (why?). To top it all off, the first request just failed with overload/quota exceeded errors (understandable, but still).
Looks like I'll wait to see if Google cares about putting the polish into a VSCode fork that at least comes close to what Cursor did.
The UI is certainly buggy, and things are getting fixed all the time. Guess it was more a "let people try the agent manager" instead of overfocusing on looks.
> Your new focus is architecting the solution, not implementing every single step. So congratulations, you have been elevated to a manager of agents.
I'm not sure many engineers will welcome this "promotion".
I don't think this is speaking to the engineers
Then who are they targeting? Who else would currently be "implementing every single step"?
Project managers and higher level management.
Reminds me of the pre-GitHub days, when I had to use CM tools designed to appeal to project and CM managers, not to the poor developers who had to use them every day. Anybody else remember Harvest?
Yeah, we'll see how that'll go.
Few horse racers became automobile racers.
If existing engineers don't change it doesn't matter because new engineers will take their place.
Horse racing didn’t go away and there are more people who race horses professionally than who race cars.
There are many more truck drivers than buggy drivers
there is a lot more buggy code than truck code
Truckers code better than bugs
"Professional riders number roughly three to six thousand worldwide, while professional drivers number roughly twenty to forty thousand across major sanctioned series."
Horses also run faster than pictures of cars
Copium
We‘ll wait and see.
Car manufacturers made profit
Some will wait and see, yes.
You weren't the target audience. The target audience was manager types tired of being told no by engineers. Always listen to the quiet parts left unspoken/unacknoeledged.
It's the same thing they tried to sell with low/no-code.
The problem is that the engineer turning what you want into code isn't normally the bottleneck. I would say about 50% of my job is helping people specify what they want sufficiently for someone to implement.
Non-technical people are used to a world of squishy definition where you can tell someone to do something and they will fill in the blanks and it all works out fine.
The problem with successful software is that the users are going to do all the weird things. All the things the manager didn't think about when they were dreaming up their happy path. They are going to try to update the startTime to the past, or to next year and then back to next week. They are going to get their account into some weird state and click the button you didn't think they could. And this is just the users that are trying to use the site without trying to intentionally break it.
I think if managers try to LLM up their dreams it'll go about as well as low/no-code. They will probably be able to get a bit further because the LLM will be willing to bolt on feature after feature and bug fix after bug fix until they realize they've just been piling up bandaids.
I am cautiously optimistic that there will be a thriving market for skilled engineers to come in and fix these things.
They will equally be tired of being told yes by LLMs.
Perhaps it's worth posing the question: what sorts of "engineers" might feel threatened by agents? Those doing engineering, or those who spend their careers wading in the shallows? Competent designers with deep comprehension, or, at best, the superficial pedants?
I just have zero faith in Google. How long until we hear that someone mysteriously got banned by Google (as we see on HN every few months? it feels like it anyway) and hear about how now they have no AI tooling etc etc etc because its all married to their Google Account.
Additionally... Google Code was shut down in 2016? I have zero confidence in such a user hostile company. They gave you a Linux phone, they extended it, and made it proprietary. They gave you a good email account, extended it and made it proprietary. They took away office software from you via Google Docs, so now you don't even own the software they do.
No thanks.
Quoting their own video:
> You can verify your code quality at a glance, then ship with absolute confidence.
Proclaiming absolute confidence after a glance leaves me with scant confident in the merit of the confidence.
I threw up a little in my mouth at "you can verify your code quality at a glance".
Why is scrolling modified on this page? I how to disable it?
Google likes to fuck with basic browser functionality for some reason. Scrolling, sometimes also how “click” intents through touch are triggered (that is, using js listeners for touch events instead of watching for the browser to communicate a “click” on an element; this does usability-killing shit like make a touch-to-stop-scrolling get interpreted as a click on whatever happens to be under your finger). I have no idea why they do this, but they do it a lot, so it must be a cultural thing.
And I don’t mean like some designers will highjack scroll to deliver a different experience like slide-like transitions or something (which may or may not be, differently, awful) but they’ll override it just to give you ordinary scrolling, except much worse (as on this page).
Seems like a lot of work to do just to make something shittier, but what do I know, I probably can’t implement a* on a whiteboard from memory or whatever.
> Google likes to fuck with basic browser functionality for some reason.
Apple is the worst offender here. Their product pages are always sluggish.
Yeah, though with theirs I can usually see why they did it (even if I’d rather they didn’t). Google’s MO is messing with scrolling for what often appears to be no user-facing reason at all.
I can't believe these "smooth scrolling" scripts are still a thing. I was wondering why I was having a hard time scrolling the page on my phone, when I got to my PC and felt the reason.
It's incredible to think how many employees of this world-leading Web technology company must have visited this site before launch, yet felt nothing wrong with its basic behavior.
Put this in your browser console to force default scrolling
Thank you. Seems the website didn't like this btw, Safari macOS. Forced return to top of page when scrolling too fast.
They want everyone to see what the webpage looks like on their Mac.
I'm on a Mac and that scrolling speed is not how Mac's scroll. The acceleration and drag are all wrong.
It's weird and laggy on Mac too. Not sure who this is for.
Came here to say this, it's super frustrating.
Google subscriptions and services are so terribly mismanaged that I will be staying away, no matter how incredible this shallow fork of vscode may be.
I remember a previous story months ago about Gemini that had Google PMs trying to hype their product, but it was all question about how nobody knows how to get Gemini API keys with any number of paid subscription.
On top of that how long until it’s https://killedbygoogle.com/ ?
Product leaders that apply world-changing technology breakthrough names to their yet-another cloned SaaS product deserve more shame.
Antigravity would be a world-changing technology. This isn't.
Why? There's nothing wrong with metaphor.
And agentic coding is about working at a much higher conceptual level. Further from the ground. Antigravity is a functional metaphor.
My only issue with it is that it's too long at five syllables, and "anti-" is an inherently negative connotation. I'm guessing this will eventually get renamed if it gets popular, much like Bard was.
They should not be too metaphorical when naming a clone, tho.
Why not? And clone of what?
As the parent said, actual anti-gravity is world-changing technology. It's telling the very laws of nature to go fuck themselves, you're gonna do what you want, even if all of known physics says it's impossible.
Working at a higher conceptual level is just project management. You're the legislator giving out unfunded mandates rather than the agency staff that has to figure out how to comply. There's power there, but it isn't anti-gravity.
That said, I suspect this is really meant to allude to https://xkcd.com/353/.
> There's power there, but it isn't anti-gravity.
That's why it's metaphor. "Operation Warp Speed" also delivered vaccines quickly, but not faster than the speed of light.
The list of company and product names that are based on a metaphor that is very obviously exaggerated is endless. Google doesn't index a googol number of pages either.
I feel it's a bit ignorant of you to double down on your argument and compare a cloned product release by some macbook swinging google engineers with vaccination, which actually positively impacted many human lives.
Are you for real? When someone disagrees, it's not "ignorance" or "doubling down". It's just legitimate disagreement. There's nothing I'm ignorant of here, so please don't throw around insults like that.
I just continue to stand by the fact that naming products using exaggerated metaphor is standard practice. The idea that it is "shameful" or "ignorant" seems absurd. I think it's OK not to take it too seriously. Nobody is going to be confused and walk off of a cliff or something because the product is named "antigravity"...
Do you get upset that the Milky Way candy bar doesn't actually contain a galaxy within? Or that the Chicago Bulls aren't as strong as actual bulls?
so your point is that you are policing what people think is shameful or ignorant? Is it too much to not think too high of a VSCode fork?
Geez, people are still this impressed by big tech?
Since when did disagreeing become policing? But yes, if someone calls me ignorant without any justification, I'm going to disagree. And if you think that's "policing", I'm sorry but you seem to be the ignorant one here around the meaning of that word.
"Geez," it's just a name. Is it too much to not get worked up over a perfectly innocent and fun name?
This 'Antigravity' software is far below the expected Google software engineering standards I have ever seen.
Quite shocking to see that Google would consider using this crass software and is the most inefficient software libraries ever made.
What were the engineers thinking?
Wait until you see Google Perpetual Motion Machine, their appointment-booking virtual assistant.
So this is Google's version of Windsurf's Wave 10 before the whole team got poached? https://windsurf.com/blog/windsurf-wave-10-browser
Trying to understand how this is anything net new in the space.
I haven't used Windsurf (been using Claude Code and similar). Does it provide an Agent Manager window/view? This to me looks more useful to me than the browser integration piece.
Looks to be. The UI has almost the exactly the same bits, and I even got 'Cascade' references as using it.
I actually like the workflow they are suggesting. There's something there for sure:
- Nano Banana => Mockup
- Antigravity/IDE => Comments/note
- Gemini => Turn to code
- Antigravity/IDE => Adjust/code
All on the same platform so can maximum automate / "agentic"
Jules
> Bajillions of dollars invested in the development of some of the most powerful computational artifacts to date.
> Fork VS Code, add a few workflow / management ideas on top.
> "Agentic development platform"
I'm Jack's depressed lack of surprise.
Please someone, make me feel something with software again.
If you want to feel something with software, leave the industry and never look back, saving programming as something you do for your own joy/reward (I'm not being hyperbolic—I'd argue we're in the early days of the web's "dark ages").
Unfortunately, once money came into the picture, quality, innovation, and anything resembling true progress flew out the window.
There is cool stuff out there! Look beyond the companies with $B valuations and you can find smart, passionate people making neat stuff.
> Please someone, make me feel something with software again.
Work with what you love, and you will never love anything again.
Challenge accepted
This is a vibe-coded VSCode fork. In a simple task, I got overload/quota exceeded errors with horrible error handling. lol
Same for me also. I have Gemini Pro subscription, still it is showing quota exceeded error.
Just from the name, I thought this was going to be Google's official take on the classic "Google Gravity" site from ages ago: https://mrdoob.com/projects/chromeexperiments/google-gravity...
I used to love leaving that site open on public PCs and watching the reactions that resulted :)
""Autonomously, an Antigravity Agent writes code for a new frontend feature, uses the terminal to launch localhost, and actuates the browser to test that the new feature works."
very interesting times; i'm glad to see browser automation becoming more mainstream as part of the ai-assisted dev loop for testing. (disclosure: started the selenium project, now working on something similar for a vibe coding context)
is that "vibium" ? as someone who tried setting up selenium in a workflow im definitely interested.
yup, vibium!
Most people are missing the point here. Testing the GUI/feature more reliable is something that Gemini 3 could unlock (looking at the ScreenSpot-Pro benchmark and its general improvement on visual understanding). At least for the (hobby-)projects I attempted this was really a bottleneck having to always test the GUI after each change as its quite often breaking something.
Oh no. What's with the scrolling in the blog page. What a terrible experience. It's clearly vibe-coded and AI-tested. If a senior saw mingling with the scrolling behavior, it would have been never in production.
> Google Antigravity is an agentic development platform, evolving the IDE into the agent-first era.
Antigravity enables developers to operate at a higher, task-oriented level by managing agents across workspaces, while retaining a familiar AI IDE experience at its core. Agents operate across the editor, terminal, and browser, enabling them to autonomously plan and execute complex, end-to-end tasks elevating all aspects of software development.
via: https://www.linkedin.com/showcase/google-antigravity/about/
I have absolutely 0 idea why any developer would rely on any IDE produced by google. It'll be canned within 5 years max, with 3-4 seeming like a reasonable estimate of the lifespan of the product
I've been using my current IDE for 17 years, and plan to continue using it for at least another 15
You mean Android Studio will be canned in 2018 max with a reasonable estimate of 2016-2017?
Isn't Android Studio based on IntelliJ and not a product developed from ground up? And Android Studio has second order revenue from the playstore.
I wouldn't be even surprised if internally the AS team's financials are counted under the Playstore umbrella.
Antigravity is based on VS Code, not designed from the ground up, and has second order revenue from the AI subscriptions (financials probably counted under the AI umbrella).
I still wouldn't trust a Google product to stick around, but these hints aren't a reliable oracle either.
When AS was launched Android was the only other viable option and it is the same even today. I don't believe Google's AI products will reach and/or sustain the same dominance as Android.
It is a product launched in the hype cycle of AI. Google has plenty of other products (launched during hype cycles) that are gathering dust.
That's not a guaranteed signal that it will meet the same fate but its something strong enough to be wary of.
It's interesting to think that Google's Antigravity is a forked version of MSFT's VS Code, which uses a browser engine built by Google, which they forked from Apple, which they forked from KHTML.
It's made by Jetbrains thankfully.
It won't matter. The core ideas of an Agent Manager view will be copied and improved by others in many project in the future.
Which is is, vi or emacs?
It will be very funny if it's vim, since Bram Moolenaar who created and ran it worked at Google from 2006 to 2021.
codeblocks. There are dozens of us!
Do you really think antigravity will last longer than 1 year?
Yes
I recognize the guys in the video, they were in marketing videos for the Windsurf IDE before its founding team was cannibalized by/absorbed into Google.
Kevin was CTO / head of product engineering at Windsurf, Anshul was a founding engineer
I found the demo videos off-putting, people come off a bit smug, not sure if intended.
I tried it. Maybe I got unlucky, but Gemini performed poorly in my testing. I gave it a task that I was working on with VSCode and GPT codex 5.1. Gemini3 repeatedly failed to finish the task and started to go down a rabbit hole on an unrelated task.
The browser extension is really cool and it provides a needed tool for the agent to use. It used the extension to show the page that it updated in the task document (the task doc is great too). However it showed me a page and did it was done, when it was clearly not done and not what I asked for.
I was expecting weaker tooling and a better model. I got good tooling and a not very good model.
Maybe 3.1 will deliver?
weird name. abbreviating it to GAG isn't much better.
Haven't we got enough of new eras yet?
No the investment amount demands a new era per week
Each VSCode fork with some random AI junk slapped on will be a new era! I can’t imagine how many eras behind I’m at this point
MacOS/Safari User here. Stuck on 'Setting Up Your Account' once I've authorized it in the browser. /shrug
To save others the trouble, it doesn't matter whether you use Chrome or Safari for the auth flow. It's broken on both. (I'm using a personal @gmail account.)
https://news.ycombinator.com/item?id=45968731
https://antigravity.google/docs/faqI love how being a paying Workspace customer for a decade or more has actually locked me out of so many Google products and features.
Not that I have any desire to try this at this point, but it's always felt ironic.
Thanks. Actually did use my personal account and got the issue nonetheless.
Same
Same
Had the same issue, have been able to sign in finally using a Google Cloud Identity (former Workspace) account by changing my IP via a VPN to Singapore. No idea why, but that worked. Tried a few other countries too, but only had success with Singapore.
I really don't know why I struggle so much with this stuff. I believe these models / agents / whatever write code that is often at least as good as the code I write, and they are super helpful tools, but it just feels like it takes away so much of the joy that is programming to me. I'm not saying it's "right" of me to feel this way, but for me the struggle, and the figuring things out by testing, identifying patterns, or looking deeper into a library's implementation (etc) is part of the challenge that makes programming and software construction fun.
I'm in your boat. I picked this career because I enjoy solving problems and thinking and understanding. I'm interested in how things work inside below the layers. To me, it's like the tech has a purpose of its own, and not just to provide value.
Using agents effectively is this whole other skillset including managing requirements, prioritization and, worse yet, I'm rarely left with any knowledge. I don't nearly get the same joy out of "I finished a task with an agent" like I do with "I had a problem, I delved deep to understand it, learned something new and solved it"
Then again, I bet people making furniture out of wood felt the same about industrial furniture factories. And it can be argued that not every use case needs custom tailored furniture...
It is kind of joyless that my day is now wait 20 min, QA, report back.
Absolutely.
This is the fruit of Windsurf brain-drain and I think it might be better than what's out there since those guys got to start from scratch from everything they learned building Windsurf
I forgot all about Windsurf from ... 4 months ago. The pace is crazy.
See the videos:
https://www.youtube.com/watch?v=YX-OpeNZYI4
https://www.youtube.com/watch?v=rKQ9b4UMpGQ
https://news.ycombinator.com/item?id=45967787
Wonder why they're unlisted.
This is common for videos embedded in a page that don't make sense out of context (e.g. in the YouTube feed).
Just sits on "Setting Up Your Account" pane and does nothing :)
Had same issue until I disconnected from Tailscale, in case that helps anyone.
I'm having the same issue. I thought it was just me or something with their cloud network. I also haven't been able to download Android Studio from the website for a month. I couldn't even download it from my Macbook so probably not the same issue.
Possibly a problem with "organization" accounts? I'm seeing the never-ending spinner too.
My personal mac worked, my work mac is spinning.
It has jamf among other stuff
Same issue here. Are you on Mac?
I am and am having the same issue. Edit: And just as I posted this comment, on my second launch, it went through after about a minute of waiting.
Tested it for roughly two hours now, far from ready for prime time, very buggy and clearly just quickly build on Windsurfs already rather issue laden code base. Essentially a less well thought out imitation of Trae's Solo mode, added in a second window on top of VSCode and not very well integrated, struggling with terminal commands and despite showing issues in the browser window and screenshot taken by the model, proclaiming the task to be completed. Tool calls also aren't as reliable as I would have expected considering their ownership of the code base, hard to tell whether that is underlying in the model (it was a major issue with 2.5 Pro) or simply Antigravity specific, hoping the latter.
Additionally, there are issues setting up accounts (Singapore VPN solved that for me), no support for Workspace users, only a free tier that requires data sharing, no additional rate limits for paying Pro or Ultra customers, etc. Even worse, Gemini CLI currently does NOT provide Gemini 3 Pro for Ultra Business customers despite paying over € 260,- per month, which is frankly ridiculous.
Will be honest, I was speculating that the reason for the multi month delay between the first A/B tests of Gemini 3 class models and the final release was so they'd have all their dugs in a row. Have some time to test everything, improve tooling, provide new paid subscriptions and/or ensure existing ones get access to everything day one, but they didn't.
Gemini 3 Pro seems very interesting (to early to say), but compared to every other recent launch by OpenAI (5, 5.1, Codex variants), Anthropic (Sonnet and Haiku 4.5), even Kimi (K2 Thinking) and Z.AI (GLM-4.6), this is by far the least organized launch of any frontier lab.
A buggy IDE which is unusable for paying customers, no CLI access for Ultra business (and none at all for Pro of any kind), etc. is frankly embarrassing when considering what competitors manage to provide the day a model launches.
What have they been working on these last two months besides going on X and posting "3" every couple of days? Why is there no paid Antigravity tier, no way to use Workspace accounts, etc? Before launching in this state, I feel it'd have been better to delay a bit more if it was absolutely needed.
Also, correct me if I'm wrong but isn't this the fourth or fifth IDE built by Google for LLM assisted coding? What happened to IDX and Firebase Studio and aren't they also based on VSCode?
Can't wait for my IDE to get sunset on me with no recourse. Thanks google, but I'll pass.
> with no recourse
Why does the IDE eat your files? If an editor shuts down, open up another one and continue. What's with the melodrama?
It's insane to me that I can't pay $20/$200 bucks after running out of limits in ~5 messages.
Why would you not at least link it to the pro and ultra accounts
at least you could upsell the pro subs to ultra. Millions of claude code and codex users who are into agentic coding is your servicable market paying attention today.
Now I'll delete antigravity and go back to codex / claude code / cursor ...
If I use this does it mean Google has access to all my code and it may popup as"AI generated" in someone else's code?
Generally if you are paying full price (paying per token), then it's not used for training.
If you are not paying, or paying a consumer level price ($20/mo) you will be trained on.
ETA: In the terms they say they use your data because "free" is the only option available in preview. However it does say you can disable sharing in your settings...
There is a setting to disable telemetry, unclear what this means though.
That's not quite how LLMs work.
Not but you can be quite sure somewhere deep inside the TOS there is a line saying their telemetry swallow your soul. If not, it will be added. It's google, that's what they do.
I can't get the agent to use my MCP server. The MCP config is provided, and the application can query the tools, but the agent can't access my server, only the http operations the application performs to list the tools is something i see in my server logs. It knows there are tools, because it sees the list, it tells me when it's trying to connect to which tool. But that fails.
where are the MCP settings in Antigravity?
Nice, if I switch now it'll be killed in two to three years right around the time Zed has all the features that I want!
On my m2 MacBook Air with 16GB of RAM, it took over 12 minutes to startup and get to a usable state. When it did, it was plainly just a jacked version of VSCode. Opening a project caused it to hang again. Dumped it. VSCodium, with the terminal pane open so I can talk to Claude works fine for me...
Was expecting https://pypi.org/project/antigravity/
We have a real, serious problem when even Google (presumably with a large share of the world's engineering might) is just forking VSCode.
Reusing open-source code is a "real, serious problem" now...?
With every AI VSCode fork we get closer to the bubble popping.
Wait till you hear about how much Linux they use.
Seems interesting, makes for the second vscode clone with ai google has made. The demo they showed in the video avoided showing code so I guess thats what they're aiming for. Although when they mentioned you can easily verify code quality by looking at end product screenshots it felt like they don't know what 'code' quality means.
The name sounds like it is not going to stick around for long.
A product name with five syllables is doomed.
"You know what I want? More Google inside my tools."
—No one, ever.
The amount of New Eras stuffed into New Eras is too damn high!
Man does this thing spam new windows... why spawn new window instead of panels in a single instance?
I have to close 4+ after just a few minutes of poking around
Vibe coding has taken a way to where its become psychological, and these days all I work with is just a "fast apt" solution like Cursor's Composer 1, I tried Gemini 3 inside Cursor, and while I had no real application to a real stratified properly to benchmark it with, it felt "already slow" in a very fast race.
It's a shame not even mention the amazing work of VSCode.
The VSCode homepage does not mention Chromium or Electron.
Read the title, got excited. Read the page.. ah well, guess we'll have to wait another while for FTL travel.
Same, I thought they had discovered cavorite with their quantum computer or something.
Long live the cursor. RIP.
I tried it and ran out of credits during the first prompt. No visible way to upgrade or purchase.
I'm stuck on "setting up this account" like most people. What a botched launch. This kind of bugginess and unreliability has become so much more frequent since big tech started tightening the screws with mass layoffs.
That's a big name for a slop fork. So many possibilities (with LLMs and without) but Google just can't bring themselves to do anything creative, let alone transformative.
They even packaged it for Linux.
Trip report: I'm using it now to revamp a dashboard I'm working on. TBH it's not feeling much better than Codex - it couldn't figure out how to launch chrome with my default profile nor how to regenerate the css with tailwind. I'm also getting a lot of model quota errors like everyone else.
I'm not switching from Claude code. It did the job but didn't do it as good as Claude Code with the details and context.
Curious if the name is a reference to https://xkcd.com/353/
> Come join us! Programming is fun again! It's a whole new world up here!
I hope Google is at least acknowledging the origin of the name, even if they are not paying royalties to Randall Monroe.
Don't take my random pondering as any kind of evidence
The name Bard didn't fly, so they went for literal flying instead.
That floating chess board was a subliminal message: Your project will teeter, and critical pieces will fall off. You will occasionally make an illegal move as the board annoyingly shifts beneath you.
can't get past the "Setting up your account" step atm
Same. I wonder if it is because I'm signing in with my Google Apps for business account.
same here, hope they fix this soon
same just hanging on the spinner
I installed it, entered one prompt, clicked the "Proceed" button, and got "Model quota limit exceeded."
Those quota limits brought me back down to earth quickly.
Especially since:
So basically just another case of vendor lock-in. No matter whether the IDE is any good - this kills it for me.>>Google Antigravity's Editor view offers tab autocompletion, natural language code commands, and a configurable, and context-aware configurable agent.
Okay, but is it configurable? Also, can you configure it to write DRY code?
Tried Antigravity for 2 queries and my model quota limit breached. Model definitely felt better than GPT 5.1 (my current daily driver). I am continuing to use Gemini 3 Pro on Cursor to evaluate further.
Did any of these VS Code forks yet fix their issues from official marketplace access leading to extensions being severely outdated and ripe with security issues?
So the whole world is a scam for your data now basically.
Neigh, a Trojan Horse.
Something something, "my kingdom for a horse." Obligatory upvote, wp
This pun made me actually laugh out loud. I almost lost some coffee.
This thing crashes on Ubuntu LTS 24.04 during start. Apparently all these agents are not able to ensure that a desktop app starts on a popular Linux distribution.
If Google has forgotten how to do Software, than the future doesn't look bright.
I enjoyed Theo’s take on this https://youtube.com/watch?v=8dTN4PBD2rg
But honestly Google software seems so buggy. The management class took over there a long ago and are quietly ruining the company.
Probably got inspired by Apple and their iOS 26 quality control
Gemini, create an RSS feed for https://antigravity.google/blog
So a VS Code fork. At this time, when VS Code is better than ever and ever improving. I think imma pass.
Why is google so bad at product branding and strategy? My complaint is aesthetic: why would you name your product a five-syllable word??
Because it's awkwardly close to the letters AGI, maybe
It's actually this, the command line tool is `agy`
Informally, people will way “antigrav”.
Its clearly a reference to the xkcd comic which does have mindshare.
Thank you for the demonstrating the reasoning that leads to these decisions.
This made me laugh, I assume you are calling me an old fogey and I will be glad to take it. This is why they don't let me near the marketing stuff.
Which comic would that be?
Later edit: Probably this one [1], which is par for the course for Alphabet, they're, conceptually, still living in the early 2010s, when this stuff was culturally relevant.
[1] https://xkcd.com/353/
well they sure seem to make product naming after going through the whole medicine cabinet too, just like the comic
I clicked around for ~30 seconds and I have no idea what this is. Great job, Google marketing
Well, I can't even use it at all, it's just stuck on the 'Setting Up Your Account' step forever. That's a pretty sad product launch lol.
Landing page instantly spun up the fan on my laptop. The animation was about 3fps.
Not sure what this gives me over Copilot in VSCode that has access to OpenAI, Claude, Grok, Gemini, and Raptor(?) models.
I assume that Copilot will have this model soon...
I enjoy that the post-login confirmation page (https://antigravity.google/auth-success) still says "Docs | Twitter"
Looks to be live but no content; OpenGraph description is "Google Antigravity - Build the new way".
I don't get how these agents can work when even Claude Sonnet 4.5 (for example) needs a lot of hand-holding for basic, simple bugfixing stuff. Wouldn't the agents just be huffing and puffing their way off the rails all the time?
They do and it is often entertaining.
This is the key question.
I downloaded it, but I was not even able to get past `Account Set Up`. :(
Not interested in trying something that will be killed shortly.
Did they build this site with Antigravity because it sure is broken on my iPad.
you would need to be bonkers to trust Google at the intersection of 1. vibe coding, 2. supporting developers, and 3. a product that's not selling ads
I know it is not place to seek help but anyone else stuck while trying changing google accounts ?
When it starts it prints: Using Cloud Code URL: https://daily-cloudcode-pa.sandbox.googleapis.com
I wonder how long they're gonna keep this around before they pull the plug.
I probably wouldn't have been drawn to coding if I was young these days based on the same motivations that led me to venture into it as a teen.
Even if this is good, I'll never know because I'm not investing time into something that will be canceled in one year.
There's no way I am using such an important piece of life as an IDE from Google just because I know they are going to kill it within 3 years, if it survives that much. Probably will die with the Windsurf guy jumping ship again.
Are they explicitly excluding OpenAI in their IDE? (gpt-oss running on Google cloud doesn't count)
No ... no, thank you. Keep your AI away from me and my stuff.
Can someone inside comment if this is this a cider fork or a new branch off vscode?
I don't want to hate on this but I remember last week, when as a backend developer doing frontend, I spent about 20 minutes prompting Claude Sonnet in a loop trying to build a landing page for a new feature.
The task was to put create a header, putting the company logo in the corner and the text in the middle.
The resulting CSS was an abomination - I threw it all away and rewrote it from scratch (using my somewhat anemic CSS knowledge), ending up with like 3 selectors with like 20 lines of styles in total.
This made me think that 1: CSS and the way we do UI sucks, I still don't get why don't we have a graphical editor that can at least do the simple stuff well. 2: when these model's don't wanna do what you want them to the way you want them, they really don't wanna.
I think AI has shown us there's a need for a new generation of simple to write software and libraries, where translating your intent into actual code is much simpler and the tools actually help you work instead of barely allowing to fight be all the accidental complexity.
We were much closer to this reality back in the 90s when you opened up a drag and drop UI editor (like VB6, Borland Delphi, Flash), wrote some glue code and out came an .exe that you could just give to people.
Somewhere along the way, the cool kids came up with the idea that GUIs are bad, and everything needs to go through the command line.
Nowadays I need a shell script that configures my typescript CDK template (with its own NPM repo), that deploys the backend infra (which is bundled via node), the database schema, compiles the frontend, and puts the code into the right places, and hope to god that I don't run into all sorts of weird security errors because I didn't configure the security the way the browser/AWS/security middleware wanted to.
>Somewhere along the way, the cool kids came up with the idea that GUIs are bad, and everything needs to go through the command line.
It's important for people to feel like "hackers" that is the primary reason why command line sort of exploded among devs. Most devs will never admit this... they may not even realize it, but I think this is the main reason it went big.
The irony is that the very thing that makes devs feel like "hackers" is the very thing that's enabling agentic AI and making developers get all resistant because they're feeling dumber.
It's 2025 and a code editor is now 600MB.
Anyone else stuck on 'setting up your account'?
Need to use a personal account. Check the first question in the FAQ: https://antigravity.google/docs/faq
My guess is it fails if you use a workspace account. I was able to use it with my personal Google account.
Hmm that does indeed seem to be the case.
Yes, its also failing on my workspace account but worked on my personal. Might be a bug or a delayed deployment for workspaces b/c it might need to be "enabled" by admins?
I'm not using a workspace account and am unable to get past this step.
Oh well. Uninstalled. This was my first experience doing software development guided by AI. Doesn't seem like a tool that will serve me well in the long run.
Doesn’t work with a workspace account for me but it does work with my private account
Is it safe to use these types of AI enhanced editors with files that contain sensitive information?
This could have been a plugin.
It's impossible to use it, stuck in the loading screen forever
Interesting concept! Curious to see how Google Antigravity works and what practical applications it might have.
I've clocked out on agentic AI IDEs. Not installing anymore until I hit an obvious wall with my current ones.
Anyone else getting this error: "Agent execution terminated due to model provider overload. Please try again later."
Yes I just tried GP3H and got this. GP3L is fine tho.
I like this tool.
edit: Scratch that, GP3L is erroring out too. Global hug I guess. I still like this.
As if this was something one could trust it won't be shut down 3 years from now...
No thanks...
Meanwhile I don't feel like the era of AI assisted software development has even started.
Nice that you can use non-Gemini models with it
I couldn't get it to work on M1 mac. it spins forever on login screen.
It's probably just under a huge load right now. Set it up pretty easily earlier on an M1 Air, but agent chats fail quickly with "model under load"
I've seen "agent" and "agentic" so many times in the last few months that the usage of the term is quickly becoming one of my biggest pet peeves. The previous one was "enshittification", and I'm glad that one was a short-lived fad.
I can't tell if you're being sarcastic about enshittification being a fading fad. I'm seeing non-technical people start to use it now.
Short lived only in my corner of reality I suppose.
I just feel second-hand embarrassment when seeing these kind of posts.
Okay, maybe I'm stupid, but the demo video included pasting and API key into the chat window.
That seems bad.
I cannot stand webpages that hijack scrolling like that.
Lame. Still like things like Zed that look a bit like vscode but without all the electron trash.
It’s just google’s attempt at cursor. Nothing to see here.
Seems like they are trying to attack both Cursor and Lovable at the same time...nice !
So deceptive name. It has nothing to do with gravity.
why isn't antigravity finishing the setup on my mac? it's been more than half an hour.
I wonder how many meetings where necessary to come up with that name
https://xkcd.com/353/
Something felt really "artificial" about that youtube video.
Really hope the quality of the IDE is better than the website...
Anyone got stuck on the "Setting Up Your Account" page ?
i tried with my personal account and still stuck on that page
yep
I switched to my personal gmail identity and it succeeded.
It's really kind of pathetic how we live in a future where "antigravity" is a text editor that lies to you, "hoverboards" are one-wheeled electric skateboards that burn your house down, and... well, can't think of a third thing at the moment, but you know the vibe.
Lotta people mining science fiction for cool names and then applying them to their crappy products, cheapening the source ideas.
A third thong could be “self driving cars” are cars that you have to stay alert and in full control of at all times.
We are in the future, it’s just a much more rubbish version than people imagined in scifi
New Era. A New - think about it for a sec - Era.
Google AI products are the new chat product
What I don't get is why they didn't fork emacs, then build a VScode mode and then add AI to that.
I guess it must have been the GPL which isn’t compatible with their AI agents.
Oh, wait I was meant to take this announcement seriously?
Nice to see that it's not locked to just Gemini models
Oh lucky, another free AI beta.
I'm going to treat this like Kiro, and just use it until they start charging for it and then probably switch back to VS code with its built-in agent support.
Eventually they're going to do a rug pull, and instead of paying $10 a month for tons of AI code request, it's going to be two or $300 for that. The economics just aren't there to actually make a profit, hopefully before the rug pool happens local models on normal hardware will be fast enough.
dupe https://antigravity.google/
I worked in a factory one summer when I was a teenager. It was a totally brain dead work, but the morale was good. The workers weren't unhappy.
I'm concerned that the new role of "manager of agents" (as google puts it) will be a soul destroying brain dead work and the morale won't be good.
yawn.
opencode with it's superior feature set and ability to use any model provider i want is....
superior
why would you even bother with google at this point?
Vim mode isn't working, oh no!
yeah I also cannot get Vim to work in Antigravity?
I get so annoyed when Google launches things like this without Vertex support. Instant no-way from compliance and I'm left sitting on the sidelines.
Neat, but the world doesn't need another IDE, and people want choice. Provide tools that plug into open workflows and step back.
Even the launch trailer sucks.
The people at Windsurf who worked on this must be laughing at us driving on their Lambos and Ferraris.
They glued slop together, shipped this and now are in Tahoe drinking Martinis watching the sunset from their private chalets.
The people that made Windsurf are in this trailer. Those who got bought by Cognition were either fired or forced to work 100h weeks
yes, hence why I said the people at Windsurf that worked on this...because they are the ones that have been shipping this slop after the buyout.
Finally! Affordable antigravity!
How long until it's killed?
I mean, google doesn't have the greatest track record.
Also, why does that site's scroll behavior is so weird? Just use the browser's default for Ford's sake!
well, I'm not even willing to try it because of whatever retarded thing they did with the page scroll
Theory: the naming of this product is strategic. Google's goal is to push something else above "Google antitrust" in the autocomplete.
It's just an xkcd reference: https://xkcd.com/353/
so it can be dumped 1 year later?
reads title
Wow was google researching some kind of anti-gravity device behind the curtains for real and then dropped it out of nowhere?
Ah damn, yet another ai-assisted-something. Crap.
Looks great, won't touch since they are probably going to do a switcheroo or a shutdown as usual.
And of course I would need to look at all the implications of spying, being locked out of google account and absence of support that are google amo. No time for that. Not for them.
What the hell is going on with scrolling on their website?
yeah... almost completely broken on my iphone. something tells me they used antigravity to vibe the website. would explain other issues mentioned like vim keybindings being ignored.
..."Youre absolutely right! I did mess up the internals of that feature and incorrectly reported that it works. let me try again..."
Notice how "Privacy" is not a talking point anywhere on that page.
⌘F only shows 1 result. and 0 in the comments here!!
A 5 syllable name for a product makes me wonder wtf the marketing team is doing.
Another Google product whose launch will be used to justify somebody's promotion, only to be left for dead only a few months later after said promoted person moves on to something else.
Why would I even bother getting mildly invested in this when the product launch/promotion incentive structure at Google is so well known?
Why is this not a VS Code Extension?
probably discontinue it in a few months
why it could not be a VS Code extension?
Program not needed.
Nice ide, can I disable the AI?
A tiny part of me was disappointed to find something other than a product involving actual antigravity.
Looking at that page makes me think I should go the other direction and switch from a graphical IDE to vim or something. You know, ground myself by adopting more gravity.
Can we mark this as a dupe of https://news.ycombinator.com/item?id=45968065
Too bad they never show these magic AI-Assisted tools being used to fix real-world bugs / implement feature requests in their open source GH repositories.
What about a demo that shows how this can be used to fix for example https://github.com/emscripten-core/emscripten/issues/24792?
My experience with GPT and Claude, is that they are fantastic for learning something brand new to me, as a kind of tutor..
But for writing code in some domain I am good in, they are pretty much useless.. I would spend a lot longer struggling to get something that barely functions from them VS writing it myself, and the one I write myself will be terse and maintainable + if it has bugs they will be like obvious ones, not insane ones that a human would never do.
Even just when getting them to write individual functions with very clear and small scopes.
Blog post: https://antigravity.google/blog/introducing-google-antigravi...
Nice demo, but they didn't say the most important thing - how much did Gemini API calls in that demo really cost? How much tokens were consumed?
I know there's a "free plan with generous rate limits" but it's obvious that they're losing money there.
Loads a blank white page and breaks the back button in Firefox.
Console error:
> Loading module from “https://antigravity.google/main-74LQFSAF.js” was blocked because of a disallowed MIME type (“text/html”).
Meanwhile I'm still on my trusty vim and letting the AI do work separately in another CLI.
I moved from Vim to Zed recently.
Any reason why this isn't just an extension instead of another fork?
I mean ok, another VSCode fork, but am I the only one seeing the .google TLD?
Oh look, another piece of shit AI slop I won't use. Next!
Another google product, there are too many and which ones will be around in a year or two?
Am I the only one stuck in the "Setting Up Your Account" loading screen?
same
Oh cool another ide for programming... aaaand its a vscode fork.
I dont know what i expected tbh
Another VSCode fork! This is getting ridiculous.
From what i saw its yet an AI first text editor. Thats a hard pass for me.
great, now everybody can vibe code even harder.
Oh no. Not another VSCode fork…
”We must also do Loveable”
Wow this page is an endless source of memes and broken UX madness
Google at its finest
Wake me up when there's an IDE that cares about performance. My coworkers are amazed my laptop can run for days because I don't use Electron crap.
and the difference from vscode is...?
This one forces you to log into your google account before you can use vscode!
VS Code won't feature in Google Graveyard in the short term future.
“agentic” is the new “ai”, which was the new “web3”, which was the new “5G”, which was the new “4G”, which was the new “HDR”, which was the new “HD”…
I don't get it. It's a completely blank web page. Did they not test in firefox?
Ah Google misconfigured their web server:
> Loading module from “https://antigravity.google/main-74LQFSAF.js” was blocked because of a disallowed MIME type (“text/html”).
Edit: And a couple minutes later, it is now working. Guess Google is reading HN.
Honestly, the full page doesn't give you much more. Not a SINGLE product image. All paragraphs about "agentic" blah-blah you have read 100s of times by now - I do not see how this is anything different from all the other AI VS Code forks, besides that it comes with Gemini from the start.
Seems like they jumped the gun on the website release, the first version I saw was a wall of text, now it has photos and videos. Maybe they have an agentic CICD
"Not a SINGLE product image."
But there is a 13 minute demo video.
https://antigravity.google/product
They're blitzing all media with a Gemini 3 push, it seems.
Looks like it's back again!
Can’t even scroll on Firefox mobile lol
Anyone want to bet how long this product will last before being killed? From looking at the frontpage, my bet is 2 years at most.
It’s…a VSCode fork? Really? What has become of Google? Ten years ago, when I was getting into the world of software, there was still an aura about them. They built everything in this huge monorepo, and it worked. They were this deeply technical company for whom it seems anything could be done.
And now they can’t even ship a desktop app without forking VSCode? Look, I get it. There’s this huge ecosystem. Everyone uses it. I’m not saying it’s damning or even bad to fork it.
But why is this being painted as something revolutionary? It’s a reskin of all the other tools which are variations on the same theme, dressed up in business speak (an agent-first UX!). I’m sure it’s OK. I downloaded it. The default Tokyo Night theme is unusable; the contrast can’t be read. I picked Vim bindings, but as soon as I tried to edit a file I noticed that was ignored.
What happened? Is this how these beautiful, innovative companies are bound to end up?
I can see why people don’t release stuff on a permissive lisense anymore. It is absolutely insane that google is even allowed to do something like this.
This. I don't think it'll move the needle. I already use vscode with copilot and it's "good enough".
They should've made an ACP server https://agentclientprotocol.com
Yeah. I really like opencode, which provides an ACP.