I was nodding along enthusiastically right up until LLMs and that point we sharply diverge.
For me, part of creating "perfect" software is that I am very much the one crafting the software. I'm learning while creating, but I find such learning is greatly diminished when I outsource building to AI. It's certainly harder and perhaps my software is worse, but for me the sense of achievement is also much greater.
I find that most of the time, programming is just procrastination, and having the LLM there breaks through that procrastination and lets me focus on the idea I was thinking on without going into the weeds.
A lot of the time, the LLM outputs the code, I test my idea, and realize I really don't care or the idea wasn't that great, and now I can move on to something else.
I hope at some point people don't feel the need to justify using or not using LLMs. If you feel like using them, use them. If you regret doing that, delete the code and write it yourself. And vice versa - if you are in a slog and an LLM can get you out, just use it.
I'm now using an LLM to write a voice note organisation application that I have been dreaming about for two decades.
I did vibe code the first version. It runs, but it is utterly unmaintainable. I'm now rewriting it using the LLM as if it were a junior or outsourced programmer (not a developer, that remains my job) and I go over every line of application code. I love it, I'm pushing out decent quality code and very focused git commits. I write every commit message myself, no LLM there. But I don't even bother checking the LLM's unit and integration tests.
I would have never gotten to this stage of my dream project without AI tooling.
Just today I gave an LLM the task of porting some Python modules to rust. I then went back and learned enough rust to understand these modules. This would have taken me days without the LLM. And I learned a lot.
I guess software is going to be like furniture - there was some (possibly imagined) period where it was very well made and bought for the long term, and (obvious where this is going) now there’s lots of ikea crap that is easy and cheap but doesn’t survive a move or getting water spilled on it, and is essentially Formica over cardboard.
Overall both are net positives, I have some nice wood furniture and also a $7 Lack bedside table, and of course I rely on some industrial long term software (Linux e.g.) but almost every day vibe code some throwaway thing for a specific task.
I like this article, I think it taps into something real.
I find myself scratching real itches that would otherwise have gone left un-scratched, because the hurdle to just getting started on something was too damn high.
For example I had need for an image contact-sheet. I'm sure there exist a lot of contact sheet generators out there, but I was able to just as quickly get claude to write me a set of scripts that took a bunch of raw images, resized them down to thumbnails, extracted their meta-data, and wrote a PDF (via Typst) with filenames and meta-data, in date order.
I got lost perfecting it, hand-picking fonts etc, but it still only took an hour or so from start to finish.
It's perfect for my need, I can customise it any time I want by simply asking claude to modify it.
Did I need to be a developer to do that? Arguably yes, to know the capabilities of the system, to know to ask it to leverage image-magick and typst, to understand what the failure-modes looked like, etc.
But I dind't need to be programmer, and over time people like the OP will learn the development side of software development without learning the programming side.
> The Extended Mind Theory argues that our tools are not just accessories, but literal extensions of our cognitive process. Viewed this way, a generic tool like a one-size-fits-all app, feels like a prosthetic that doesn’t quite fit.
Three years back now, my wife and I were planning a two-week long trip and found it really difficult to simply move day-places around (e.g. shift a whole day in a schedule with all of the places planned for the day) as we were planning a multi-city route.
We started with Google Sheets (way too cumbersome), then Docs (cumbersome in a different way), then a simple app using Firebase + the Google Maps embedded API built over a weekend, and then ended up building a full blown planning app and eventually a Chrome extension[0] that hooks directly into Google Maps (our preferred tool for exploring).
We are meticulous planners so I totally get the author's sentiment here. Many people see the app the first time and feel overwhelmed, but for us, it's hard to imagine using other tools now because this one fits "just right" having been built specifically for our planning process.
What we really need is well-built open source software base with flexible and well-documented plugin SDKs where people can just vibecode their own extensions on top of them. The software needs to be designed from scratch expressly for the purpose of being extended in this way, and the SDK should place appropriate limits to allow multiple extensions to be combined easily.
Otherwise, there is too much you have to do right before you have a suitable software base to start building your extra personalized features on. Building on existing open-source software (not designed to be extended on) isn't great either because you would need to merge any changes from the original software into your fork, as opposed to a purpose-built SDK that would better tolerate plugins on different base software versions.
I'm working on this for gaming but the idea is really applicable to any kind of software, if the goal is to allow people to easily create and run personalized versions of them without as much effort and chance for things to go wrong.
When I saw "Perfect Software" in the title, I thought it referred to Perfect Software, the developer who produced the Perfect Writer word processor, Perfect Calc spreadsheet, and Perfect Filer database. These were a suite of office software products developed in the early '80s for CP/M and MS-DOS computers.
However, I'm skeptical about AI, because what I've understood about agentic processes is more about cheap dopamine.
When it comes to medium-sized software development (over 50k LOC), there is much less fun and much more pain, because a growing codebase doesn't allow you to make new features easily.
I believe it is important not to mix up a dose of dopamine gotten from agentic results, as in the article, with achievement from longstanding work, even if it's not so attractive from a short-term perspective.
I use Claude Code on a 900kloc Rails/JS monolith and it’s still pretty pleasant. However if it wasn’t already structured well, I could see that being a worse experience.
The article is about the pleasure of creating new software (occasionally with AI help), so I hope that the 900k LOC in your case doesn't come from Claude Code.
Someone just discovered why some of us enjoy shell scripts and Emacs so much. No need for LLM, just small hacks to solve specific problems in limited contexts that works well enough, most of the time, for one user.
Even phones these days have pretty good scripting environments (thinking of Shortcuts on iOS).
I’ve had really good luck with Claude helping me write shortcut scripts. For example, I wore a CGM this year for a bit and I couldn’t find an easy way to get the raw data. It did log everything to Apple Health and a shortcut was able to extract it and append it to a spreadsheet where each row was a reading.
About 15 years ago I thought of writing custom software for friends and family instead of paying for gifts in order to save money (and give something more “meaningful”). For instance a fun guessing game using photos from a group holiday.
I never did it because I imagined the pain of supporting every device or screen size, or dealing with someone who wants to know why their gift stopped working 6 months later.
The gains I’ve seen from LLM code - making me personally more productive in languages I’ve never used before - don’t erase the support burden so I think I’d still avoid this.
Still, I wonder if soon ordinary people will find it easy enough to make software for their own amusement (not just us nerds doing side projects to stay current), and will my job ever morph into being a software “mechanic“ who is paid to fix what someone else built? Not just “someone else working at the company who owns the software”, but a different company or individual entirely?
Will software maintenance become the job that big industry stops wanting to take because it’s so cheap to write something new that they’ll never fix this year’s model?
Or is software maintenance being democratised by LLMs such that a corner software shop could realistically offer maintenance for this one copy of a piece of software on this one device that the customer brings in?
I think we’ve never discussed a “software right to repair” because changing software is expensive, but we might see that change.
What seems perfect is the adaptability & changeability, the suiting to fit.
We see LLMs as a huge opportunity here, to self define.
And existing software as too limberous & weighty.
But there are so many other dimensions and levels of how and why the past hasn't let us situate our software and us together effectively. The architecture of software doesn't have this grow-in-ability to it.
I love the Home-cooked Software and Barefoot Developers material. But neither of those ideas nor perfect software nor audience of one actually appeal to me that strongly. They are all very positive enormous breaks from bad software and bad times where we didn't have basic liberty over systems. But they all strike me as valorizing a particularly isolated rejectionist view of software, that ultimately is rude to the protocols & affordances building that a good healthy and connected form of software that we might and perhaps SHOULD aspire to.
But anything unjamming is from the inflexible unarticulated illegible mess of systems we can at best endure today is doing great work. Many positive steps into greater beyonds out of bad tar pits. 2025 has amazing hope amid all this.
This resonates with me. I'm not a programmer, and before LLM's I could only make basic hello world apps and simple websites. Now I am developing my own versions of various apps that I've used but maybe have limitations that I've become frustrated with. For example, I didn't like how the fitness tracker Strava didn't allow me to customize audio announcements, so now I have my own (and in my own eyes) better version of Strava that I use instead. It's absolutely blowing my mind that this is possible and available today, and not some tech-optimists wet dream about an impossible future.
By the authors definition, I've been writing perfect software for over a decade.
It's never required LLMs. In fact, I think the idea that "LLMs allow us to write software for ourselves" borders on missing the point, for me at least. I write software for myself because I like the exploratory process .. figuring out how do do something such that it works with as little friction as possible from the side of the user; who is of course myself, in the future.
I like nitpicking the details, getting totally side-tracked on seemingly frivolous minutiae. Frequently enough, coming to the end of a month long yak-shave actually contributes meaningful insight to the problem at hand.
I guess what I'm trying to say is "you're allowed to just program .. for no other reason than the fun of it".
As evidence for my claims: a few of my 'perfect' projects
I get what you're saying - I personally scratch that itch by doing woodworking and hobby electronics; I just love doing it and the end product is often just a means to an end; to craft something and enjoying the process of it.
But programming doesn't give me that same feeling, and honestly; the scope of doing and learning everything needed to make my projects without LLM's are just way out of reach. Learning these things would not be relevant to my career or my other hobbies. So, for me I use LLM's the way a person who's not into carpentry might buy the services of a carpenter, despite the possibility of them doing the project themselves after investing tons of time into learning how.
These days, I spend my personal coding time on building personal interfaces either as a shell script or as emacs packages. So many tools and applications hinders power usage.
I’ve always been somewhat dissatisfied with image viewing/browsing software. Gqview and Sequential came close to being what I wanted, but there were things about both that I didn’t like. I finally just wrote my own custom viewer using pyqt. For me, it is perfect software!
I see it as just another tool. Like any other tool, you have to learn it and integrate it into your workflow.
And as a learning tool, it’s extraordinary.
Not because it replaces understanding, but because it accelerates it: you can explore unfamiliar domains, compare approaches, and iterate with feedback that used to take days or weeks.
The responsibility to think, judge, and decide still sits entirely with the developer.
But as for today, have we all just collectively decided to pretend that the LLMs we have are capable of writing good software?
I use LLMs a lot in my workflow, I probably spend a whole day per week learning and fiddling with new tools, techniques, etc. and trying to integrate them in all sorts of ways. Been at it for about a year and a half, mainly because I’m intrigued.
I’m sorry but it still very much sucks.
There are things it’s pretty good at, but writing software, especially in large brownfield projects, is not one of them. Not yet, at least.
This article triggers all my “written by an LLM” spidey senses.
Which is ironic considering the subject matter. “Perfect”, but artificially constructed. “Just for me”, but algorithmic slop.
I agree that you can do so much more custom tailoring of bespoke software with the speed an LLM brings. But something inside of me still revolts at calling this anything other than “convenient”.
“Perfect” I will reserve for things I’ve made myself. However “imperfect” they may really be.
Look at it this way. The carver doesn't have to grow the tree. Using an LLM for coding is a lot like being a carver. You can take broad or small strokes and discard what you don't like.
This has always been possible, even if you weren't a programmer. You just needed to have the desire to customize your computing environment, and the time and patience to do it.
There is so much software out there, written by people who wanted to solve their particular problem, just like you. Chances are that some of it will fit your needs, and, if the software is flexible enough, will allow you to customize it to make that fit even better.
This is why the Unix philosophy is so powerful. Small independent programs that do one thing well, which can be configured and composed in practically infinite number of ways. I don't need to write a file search or sorting program, nor does the file search program need to implement sorting. But as a user, I can compose both programs in a way to get a sorted list of files by any criteria I need. This is possible without either program being aware of the other, and I can compose programs written decades ago with ones written today.
You can extend this to other parts of your system as well. Instead of using a desktop environment like GNOME, try using a program that just manages windows. Then pick another program for launching applications. And so on. This is certainly more work than the alternative, but at the end of the day, you feel like you are in control of your computer, instead of the other way around.
I was nodding along enthusiastically right up until LLMs and that point we sharply diverge.
For me, part of creating "perfect" software is that I am very much the one crafting the software. I'm learning while creating, but I find such learning is greatly diminished when I outsource building to AI. It's certainly harder and perhaps my software is worse, but for me the sense of achievement is also much greater.
I find that most of the time, programming is just procrastination, and having the LLM there breaks through that procrastination and lets me focus on the idea I was thinking on without going into the weeds.
A lot of the time, the LLM outputs the code, I test my idea, and realize I really don't care or the idea wasn't that great, and now I can move on to something else.
I hope at some point people don't feel the need to justify using or not using LLMs. If you feel like using them, use them. If you regret doing that, delete the code and write it yourself. And vice versa - if you are in a slog and an LLM can get you out, just use it.
I love this answer
I really do
I'm now using an LLM to write a voice note organisation application that I have been dreaming about for two decades.
I did vibe code the first version. It runs, but it is utterly unmaintainable. I'm now rewriting it using the LLM as if it were a junior or outsourced programmer (not a developer, that remains my job) and I go over every line of application code. I love it, I'm pushing out decent quality code and very focused git commits. I write every commit message myself, no LLM there. But I don't even bother checking the LLM's unit and integration tests.
I would have never gotten to this stage of my dream project without AI tooling.
You can still create what you already know how to, by hand, but also extend to areas you previously where shy about with the help of LLMs.
Just today I gave an LLM the task of porting some Python modules to rust. I then went back and learned enough rust to understand these modules. This would have taken me days without the LLM. And I learned a lot.
Are you writing software for the sense of accomplishment or to create software you wish you had?
The two aren't mutually exclusive.
IMO LLMs/AI alone neither make nor break anything.
I guess software is going to be like furniture - there was some (possibly imagined) period where it was very well made and bought for the long term, and (obvious where this is going) now there’s lots of ikea crap that is easy and cheap but doesn’t survive a move or getting water spilled on it, and is essentially Formica over cardboard.
Overall both are net positives, I have some nice wood furniture and also a $7 Lack bedside table, and of course I rely on some industrial long term software (Linux e.g.) but almost every day vibe code some throwaway thing for a specific task.
I like this article, I think it taps into something real.
I find myself scratching real itches that would otherwise have gone left un-scratched, because the hurdle to just getting started on something was too damn high.
For example I had need for an image contact-sheet. I'm sure there exist a lot of contact sheet generators out there, but I was able to just as quickly get claude to write me a set of scripts that took a bunch of raw images, resized them down to thumbnails, extracted their meta-data, and wrote a PDF (via Typst) with filenames and meta-data, in date order.
I got lost perfecting it, hand-picking fonts etc, but it still only took an hour or so from start to finish.
It's perfect for my need, I can customise it any time I want by simply asking claude to modify it.
Did I need to be a developer to do that? Arguably yes, to know the capabilities of the system, to know to ask it to leverage image-magick and typst, to understand what the failure-modes looked like, etc.
But I dind't need to be programmer, and over time people like the OP will learn the development side of software development without learning the programming side.
And that's okay.
We started with Google Sheets (way too cumbersome), then Docs (cumbersome in a different way), then a simple app using Firebase + the Google Maps embedded API built over a weekend, and then ended up building a full blown planning app and eventually a Chrome extension[0] that hooks directly into Google Maps (our preferred tool for exploring).
We are meticulous planners so I totally get the author's sentiment here. Many people see the app the first time and feel overwhelmed, but for us, it's hard to imagine using other tools now because this one fits "just right" having been built specifically for our planning process.
[0] For anyone interested: https://chromewebstore.google.com/detail/turasapp/lpfijfdbgo...
What we really need is well-built open source software base with flexible and well-documented plugin SDKs where people can just vibecode their own extensions on top of them. The software needs to be designed from scratch expressly for the purpose of being extended in this way, and the SDK should place appropriate limits to allow multiple extensions to be combined easily.
Otherwise, there is too much you have to do right before you have a suitable software base to start building your extra personalized features on. Building on existing open-source software (not designed to be extended on) isn't great either because you would need to merge any changes from the original software into your fork, as opposed to a purpose-built SDK that would better tolerate plugins on different base software versions.
I'm working on this for gaming but the idea is really applicable to any kind of software, if the goal is to allow people to easily create and run personalized versions of them without as much effort and chance for things to go wrong.
So, something like Excel?
Very OT, and this betrays my age, but...
When I saw "Perfect Software" in the title, I thought it referred to Perfect Software, the developer who produced the Perfect Writer word processor, Perfect Calc spreadsheet, and Perfect Filer database. These were a suite of office software products developed in the early '80s for CP/M and MS-DOS computers.
Here is the demo of my perfect software.
https://play.tirreno.com
However, I'm skeptical about AI, because what I've understood about agentic processes is more about cheap dopamine.
When it comes to medium-sized software development (over 50k LOC), there is much less fun and much more pain, because a growing codebase doesn't allow you to make new features easily.
I believe it is important not to mix up a dose of dopamine gotten from agentic results, as in the article, with achievement from longstanding work, even if it's not so attractive from a short-term perspective.
I use Claude Code on a 900kloc Rails/JS monolith and it’s still pretty pleasant. However if it wasn’t already structured well, I could see that being a worse experience.
The article is about the pleasure of creating new software (occasionally with AI help), so I hope that the 900k LOC in your case doesn't come from Claude Code.
Someone just discovered why some of us enjoy shell scripts and Emacs so much. No need for LLM, just small hacks to solve specific problems in limited contexts that works well enough, most of the time, for one user.
Even phones these days have pretty good scripting environments (thinking of Shortcuts on iOS).
I’ve had really good luck with Claude helping me write shortcut scripts. For example, I wore a CGM this year for a bit and I couldn’t find an easy way to get the raw data. It did log everything to Apple Health and a shortcut was able to extract it and append it to a spreadsheet where each row was a reading.
About 15 years ago I thought of writing custom software for friends and family instead of paying for gifts in order to save money (and give something more “meaningful”). For instance a fun guessing game using photos from a group holiday.
I never did it because I imagined the pain of supporting every device or screen size, or dealing with someone who wants to know why their gift stopped working 6 months later.
The gains I’ve seen from LLM code - making me personally more productive in languages I’ve never used before - don’t erase the support burden so I think I’d still avoid this.
Still, I wonder if soon ordinary people will find it easy enough to make software for their own amusement (not just us nerds doing side projects to stay current), and will my job ever morph into being a software “mechanic“ who is paid to fix what someone else built? Not just “someone else working at the company who owns the software”, but a different company or individual entirely?
Will software maintenance become the job that big industry stops wanting to take because it’s so cheap to write something new that they’ll never fix this year’s model?
Or is software maintenance being democratised by LLMs such that a corner software shop could realistically offer maintenance for this one copy of a piece of software on this one device that the customer brings in?
I think we’ve never discussed a “software right to repair” because changing software is expensive, but we might see that change.
I feel the same, I'm now building more side projects with the help of AI even if they're only for me
What seems perfect is the adaptability & changeability, the suiting to fit.
We see LLMs as a huge opportunity here, to self define.
And existing software as too limberous & weighty.
But there are so many other dimensions and levels of how and why the past hasn't let us situate our software and us together effectively. The architecture of software doesn't have this grow-in-ability to it.
I love the Home-cooked Software and Barefoot Developers material. But neither of those ideas nor perfect software nor audience of one actually appeal to me that strongly. They are all very positive enormous breaks from bad software and bad times where we didn't have basic liberty over systems. But they all strike me as valorizing a particularly isolated rejectionist view of software, that ultimately is rude to the protocols & affordances building that a good healthy and connected form of software that we might and perhaps SHOULD aspire to.
But anything unjamming is from the inflexible unarticulated illegible mess of systems we can at best endure today is doing great work. Many positive steps into greater beyonds out of bad tar pits. 2025 has amazing hope amid all this.
This resonates with me. I'm not a programmer, and before LLM's I could only make basic hello world apps and simple websites. Now I am developing my own versions of various apps that I've used but maybe have limitations that I've become frustrated with. For example, I didn't like how the fitness tracker Strava didn't allow me to customize audio announcements, so now I have my own (and in my own eyes) better version of Strava that I use instead. It's absolutely blowing my mind that this is possible and available today, and not some tech-optimists wet dream about an impossible future.
By the authors definition, I've been writing perfect software for over a decade.
It's never required LLMs. In fact, I think the idea that "LLMs allow us to write software for ourselves" borders on missing the point, for me at least. I write software for myself because I like the exploratory process .. figuring out how do do something such that it works with as little friction as possible from the side of the user; who is of course myself, in the future.
I like nitpicking the details, getting totally side-tracked on seemingly frivolous minutiae. Frequently enough, coming to the end of a month long yak-shave actually contributes meaningful insight to the problem at hand.
I guess what I'm trying to say is "you're allowed to just program .. for no other reason than the fun of it".
As evidence for my claims: a few of my 'perfect' projects
https://github.com/scallyw4g/bonsai
https://github.com/scallyw4g/poof
https://scallywag.software
I get what you're saying - I personally scratch that itch by doing woodworking and hobby electronics; I just love doing it and the end product is often just a means to an end; to craft something and enjoying the process of it.
But programming doesn't give me that same feeling, and honestly; the scope of doing and learning everything needed to make my projects without LLM's are just way out of reach. Learning these things would not be relevant to my career or my other hobbies. So, for me I use LLM's the way a person who's not into carpentry might buy the services of a carpenter, despite the possibility of them doing the project themselves after investing tons of time into learning how.
These days, I spend my personal coding time on building personal interfaces either as a shell script or as emacs packages. So many tools and applications hinders power usage.
I’ve always been somewhat dissatisfied with image viewing/browsing software. Gqview and Sequential came close to being what I wanted, but there were things about both that I didn’t like. I finally just wrote my own custom viewer using pyqt. For me, it is perfect software!
I disagree.
How do we work together when we all have our own unilateral views of software?
I see it as just another tool. Like any other tool, you have to learn it and integrate it into your workflow.
And as a learning tool, it’s extraordinary. Not because it replaces understanding, but because it accelerates it: you can explore unfamiliar domains, compare approaches, and iterate with feedback that used to take days or weeks.
The responsibility to think, judge, and decide still sits entirely with the developer.
Great vision for 2030.
But as for today, have we all just collectively decided to pretend that the LLMs we have are capable of writing good software?
I use LLMs a lot in my workflow, I probably spend a whole day per week learning and fiddling with new tools, techniques, etc. and trying to integrate them in all sorts of ways. Been at it for about a year and a half, mainly because I’m intrigued.
I’m sorry but it still very much sucks.
There are things it’s pretty good at, but writing software, especially in large brownfield projects, is not one of them. Not yet, at least.
I’m starting to believe many are just faking it.
This article triggers all my “written by an LLM” spidey senses.
Which is ironic considering the subject matter. “Perfect”, but artificially constructed. “Just for me”, but algorithmic slop.
I agree that you can do so much more custom tailoring of bespoke software with the speed an LLM brings. But something inside of me still revolts at calling this anything other than “convenient”.
“Perfect” I will reserve for things I’ve made myself. However “imperfect” they may really be.
Look at it this way. The carver doesn't have to grow the tree. Using an LLM for coding is a lot like being a carver. You can take broad or small strokes and discard what you don't like.
This has always been possible, even if you weren't a programmer. You just needed to have the desire to customize your computing environment, and the time and patience to do it.
There is so much software out there, written by people who wanted to solve their particular problem, just like you. Chances are that some of it will fit your needs, and, if the software is flexible enough, will allow you to customize it to make that fit even better.
This is why the Unix philosophy is so powerful. Small independent programs that do one thing well, which can be configured and composed in practically infinite number of ways. I don't need to write a file search or sorting program, nor does the file search program need to implement sorting. But as a user, I can compose both programs in a way to get a sorted list of files by any criteria I need. This is possible without either program being aware of the other, and I can compose programs written decades ago with ones written today.
You can extend this to other parts of your system as well. Instead of using a desktop environment like GNOME, try using a program that just manages windows. Then pick another program for launching applications. And so on. This is certainly more work than the alternative, but at the end of the day, you feel like you are in control of your computer, instead of the other way around.
The comparison to coffee is bad, since it's obvious the best coffee is with no sugar and no milk.
I agree, but it is there: perfect amount of sugar: 0, perfect milk-to-coffee ratio: 0 to 1.
What really gets on my nerves is the justified text...