I would like to see Raku (https://raku.org) at least mentioned in the list of languages to be aware of. Why?
- Raku has built in Grammars so it is a great place to do early iteration of your parser
- Raku is objects and type classes all the way down (as explained here https://gist.github.com/raiph/849a4a9d8875542fb86df2b2eda89296 )
- RakuAST development is well advanced (use v6.e.PREVIEW) with the Slangify module to accelerate development of sub languages (Slangs)
The main idea of renaming from Perl6 to Raku was to allow this beautiful and seductive new language to escape the black hole gravity well formed by the collapse of the Perl star. Seems like Raku is stuck inside the Perl event horizon for ever, with no hope of reputational escape.
That's nice summary of the space and how large it is. My recommendation is to just start with math expression parser and evaluator. You can start with Pratt but I would even recommend going with infix to reverse polish using stack.
Adding construct like IF or variables is naturally next step but you will have code in place and idea where to put it and how approach it.
I learned a lot about JVM runtime, how Zig is parsing itself, how Lua represents values... Too many good rabbit holes to fall in.
I think you should probably start by asking yourself if you should design a new language. Most new languages fall in the bucket of low value innovation that is instant tech debt for anyone who tries to use it for real
Even the successful ones are often pointless variations on a theme. Ruby, perl & python don't all need to exist for example, as they essentially do the same thing, about as poorly. Now python has won we should just drop the others
Different languages excel at different things. There shouldn’t be a “one size fits all” otherwise we’d be writing software in FORTRAN and assembly.
And designing a language is a good exercise if purely from an academic perspective. Eg you learn how to write parsers, and a bunch of PL theory that we take for granted when just being a consumer of a programming language.
Not everything needs to be done with global domination in mind.
I started programming assembly in 2025 for 6592 and Z80 cpus and believe me: it is fun and IMO actually easier then lets say learning Haskell or JS from scratch.
You’re missing my point. I’m not criticising FORTRAN nor assembly. I’m saying that people wouldn’t have created C, Java, Python, Pascal, BASIC, and so on and so forth if everyone said “why bother creating a new language when we already have something perfectly good here”
You missed his point. He's not saying "why bother why a new language if this one is fine", he's saying "why bother with a very similar language if this one is fine".
I think that's fair. Even if you are just doing a hobby language there are plenty of unexplored niches, e.g. that compile-to-shell language I've forgotten the name of.
I didn’t miss his point. Back in the 70s, many of the new languages were just a subtle variation on the previous one. It’s only later they evolved into something distinctive.
Which is why I said we’d still be using FORTRAN.
Languages that start out radically different don’t tend to gain momentum. Whereas languages that are familiar tend to grow and introduce new ideas.
Nothing is invented in a vacuum.
Also I completely disagree that one shouldn’t create a hobby project need to be innovative. Sometimes people do create things just because they can. And it’s a good thing too because otherwise we wouldn’t have half the open source software available to us today. Many of which was originally intended for personal use, including Linux.
The problem these days is we’re so brainwashed by stories of unicorn start ups pumped with VC money that now everyone thinks every hobby needs to has a viable business plan underneath. It’s like people have forgotten how to play for fun.
So people should go out and create new programming languages. The worst that would happen is they learn to be a better programmer in their day to day language.
I believe there are far more interesting stuff to learn about these languages, like the whole category of runtimes could have been mentioned, which can directly affect the language design itself (e.g. having GC vs some language feature for managing memory, open vs closed world model, having an async feature in the language or let the runtime handle it, etc)
I’m waiting for a llm focused language. We’re already seeing AI is better with strongly typed languages. If we think about how an agent can ensure correctness as instructed by a human, as the priority, things could get interesting. Question is, will humans actually be able to make sense of it? Do we need to?
It’s basically called “reinforced learning” and it’s a common technique for machine learning.
You provide a goal as a big reward (eg test passing), and smaller rewards for any particular behaviours you want to encourage, and then leave the machine to figure out the best way to achieve those rewards through trial and error.
After a few million attempts, you generally either have a decent result, or more data around additional weights you need to apply before reiterating on the training.
Defining the goal is the easy part: as I said in my OP, the goal is unit tests passing.
It’s the other weights that are harder. You might want execution speed to be one metric. But how do you add weights to prevent cheating (eg hardcoding the results)? Or use of anti-patterns like global variables? (For example. Though one could argue that scoped variables aren’t something an AI-first language would need)
This is where the human feedback part comes into play.
It’s definitely not an easy problem. But it’s still more pragmatic than having a human curate the corpus. Particularly considering the end goal (no pun intended) is having an AI-first programming language.
I should close off by saying that I’m very skeptical that there’s any real value in an AI-first PL. so all of this is just a thought experiment rather than something I’d advocate.
Yes. The learning comes from running tests on the program and ensuring they pass. So running as an agent. Tests and compiler give hard feedback- thats the data outside the model that it learns from.
I think modern RLHF schemes have models that train LLMs. LLMs teaching each other isn't new.
Interesting page. The latest language I designed is an stack based intermediate language for a C compiler. Not realy intended for human usage, but readable in the sense that you can compare it with the original C code.
Ha! I just wrote stack based RISC cpu architecture with assembler and now thinking about implementing my own FORTH like lang (niche stack based programming language) compiler.
- this new language looks and behaves exactly like javascript, except it doesnt have "eval" and "new Function", so it is CSP safe. That's the only difference. I wanted to execute dynamically generated code in chrome extension
- llm did most of the work of creating a nearley grammar and associated interpreter (whole thing is bundled, nearley is not a final dependency), elaborate tests make this quite sane to handle
- took me about total of 1 weeks for the initial mvp to try out, and then have been fixing bugs and inconsistencies with javascript behavior, about 1 day a month of effort
- mostly 0
The only reason to create was I couldnt find something similar and it was low effort thanks to llm
I also created another even smaller DSL you can say
Nice. I built something basically just like this for work for the same reason last year. It only look a few hours though, cause I just used Acorn [0] to parse my JS, then directly evaluated the AST. It also had an iteration limit and other configurable limits so I can eval stuff in the browser without crashing the tab. I did not use an LLM.
csp safe has a particular meaning associated it with. its not a "safe" language whatever that is. chrome webstore team is okay with it and serves my purpose. if you have submitted extensions to google chrome then you would know that any sign of "eval" or new Function in the code will lead to rejection.
I want to execute dynamically generated javascript looking code in chrome extension without using eval or new function. basically eval without actually using eval.
linter would help me find and avoid usages of eval.
I think most popular languages were started as an experiment in some feature, or to solve a specific problem someone had. Those are good reasons to make a language. I see no reason to make a language just to take attention away from other existing languages. Instead, make a language so you can understand how to make languages. It is 100% doable by one person. It's fun and educational.
My idea of for fun is to release something people will use. I have more fun if I work on something useful. For me is less the journey than the end goal.
I love working on software, architecture, design but only if I see some use.
Of course, for other people, the journey is more interesting than the destination and they have fun hacking stuff just for the sake of it. They discover things and learn new stuff they wouldn't have learned otherwise. And this is a path at least as valid as the other.
1. Yes, as long as there are new machines that need programming, new programming languages will be needed. Today's top languages were built for the machines of the 1970, 80s, and 90s. Tomorrow's languages will be built for machines of today and tomorrow. As Alan Kay put it, if you want to invent a new language, first invent the machine of the future and then build a language for it.
2. No, you cannot. First of all, PL devs are cats, it's very difficult collecting them without financial compensation. So if your plan is to post a language and hope that people will come help you, you'll likely be disappointed. The problem is that everyone else interested in building PLs has their own itch to scratch, and they're not going to scratch yours without some compensation.
You might think "Well I can just raise money to do this", and you would be wrong. First, it's very hard to raise money for PLs. Usually you have to have come sort of cred to do it. I know of only 3 projects to have raised VC money for a PL project, and they each had some success before they had done so: Chris Granger (Light Table), Paul Biggar (CircleCI), and Chris Lattner (Swift/LLVM). Granger's project Eve raised $2M and ran out of money after 3 years; Biggar's project Dark also raised money, then fired all the devs when he realized he was burning cash too fast, then he slow-burned development for years, then he gave up and handed development over to someone else; and Lattner raised almost $100M for Mojo, which is probably going to end much the same way as Eve and Dark, but I wish them the best.
Anyway, the point is that you personally (no offense) don't have the profile to raise $100M like Lattner. $2M is not enough for a PL project. Lattner is keeping Mojo closed source for now because there's no good answer for how they're going to make enough money as an open source language to justify raising $100M.
And the reason it's so hard to raise money is because there's no money to be made. No one pays for PLs. No one pays for PL dev tools. They have to be open source or they're rejected by the dev community. The only ones these days who can reasonably pay for all of this with no potential revenue stream are giant corporations, who use the lang as a hook into their ecosystem.
3. Even though the answer is no, you yourself can still get an MVP off the ground in a pretty reasonable amount of time. It's never been easier to make a PL. The problem with PLs is building them is kind of like measuring the coastline; language projects are fractals -- there's an infinite amount of detail you can work on in any given direction. It's very easy for a language project to become a language + editor project, and it's easy for that to turn into language + editor + operating system if you're not disciplined. Plenty of PL devs have fallen into that trap.
4. Rounds to 0% chance. You'll be lucky if you build something that even you will use. Rather, most PL devs end up working on their language in some other language, because working on languages is what they want to do!
That said, it's still important to write languages that you understand no one will use. First it allows you to try new things that may good but unpopular. If PL devs only did what was popular with devs, PLs would go nowhere as a field.
Consider the so called "Hornet's nest" of programming languages [1], which is the tightly related cluster of imperative programming languages which have been the most researched and used over the last 50 years. There is a vaaaaaaaast design space outside that nest, begging for more language development. No one will use most of them, but it's important to understand what those languages might look like to maybe find some new ideas that work.
Also "didn't make it" is kind of an unfair judgement. Gaining popularity doesn't have to be a goal. In fact, it shouldn't be a goal if you want to have any fun at all. There's an infinite amount of work to be done, and if you're not doing it for you, you won't get far at all. That's really the only way to fail at this.
Bold plus, making PLs is a lifestyle, not a business. Most PLs clones each other and absorb features. The only difference is QOL and tooling. Users expect to have a full set of batteries, an IDE/LSP, jobs, OOP style, and minimal effort to learn. Being popular contradicts with the idea of pushing the boundaries and shifting paradigms.
I think at this point we have to assume all articles are attempts to make some llm better, so they're spewing this shit out and if it gets a genuine reaction, we just gave the llm a rf-cookie. And if someone says "crap llm post" then it gets punished.
I would like to see Raku (https://raku.org) at least mentioned in the list of languages to be aware of. Why?
Here is a Raku implementation of Brainfuck to whet the appetite https://github.com/alabamenhu/PolyglotBrainfuck/blob/main/li...For some reason https://gist.github.com/raiph/849a4a9d8875542fb86df2b2eda892... wasn't a link in your comment but it was a great read
For those unaware, Raku is the evolution of Perl 6, basically. It's honestly a beautiful and seductive language. At the same time it terrifies me.
The main idea of renaming from Perl6 to Raku was to allow this beautiful and seductive new language to escape the black hole gravity well formed by the collapse of the Perl star. Seems like Raku is stuck inside the Perl event horizon for ever, with no hope of reputational escape.
Great analogy, and similar to how I saw things play out.
IIRC Perl 6 wanted to expand or morph into something better, spent a ton of time on it, and the community in general rejected it hard.
So now we have this dangling language that's shunned by its own community, regardless of its merits. Weird place to be in.
lol
Well Raku is not shunned by the very warm and welcoming Raku Community … https://raku.org/community
https://forum.rescript-lang.org/t/introducing-moonbit-and-a-... I happend to write a post sharing experience in building Rescript and MoonBit language
That's nice summary of the space and how large it is. My recommendation is to just start with math expression parser and evaluator. You can start with Pratt but I would even recommend going with infix to reverse polish using stack.
Adding construct like IF or variables is naturally next step but you will have code in place and idea where to put it and how approach it.
I learned a lot about JVM runtime, how Zig is parsing itself, how Lua represents values... Too many good rabbit holes to fall in.
I think you should probably start by asking yourself if you should design a new language. Most new languages fall in the bucket of low value innovation that is instant tech debt for anyone who tries to use it for real
Even the successful ones are often pointless variations on a theme. Ruby, perl & python don't all need to exist for example, as they essentially do the same thing, about as poorly. Now python has won we should just drop the others
Terrible advice.
Different languages excel at different things. There shouldn’t be a “one size fits all” otherwise we’d be writing software in FORTRAN and assembly.
And designing a language is a good exercise if purely from an academic perspective. Eg you learn how to write parsers, and a bunch of PL theory that we take for granted when just being a consumer of a programming language.
Not everything needs to be done with global domination in mind.
FYI:
I started programming assembly in 2025 for 6592 and Z80 cpus and believe me: it is fun and IMO actually easier then lets say learning Haskell or JS from scratch.
Assemblers with macros are amazingly simple.
You’re missing my point. I’m not criticising FORTRAN nor assembly. I’m saying that people wouldn’t have created C, Java, Python, Pascal, BASIC, and so on and so forth if everyone said “why bother creating a new language when we already have something perfectly good here”
You missed his point. He's not saying "why bother why a new language if this one is fine", he's saying "why bother with a very similar language if this one is fine".
I think that's fair. Even if you are just doing a hobby language there are plenty of unexplored niches, e.g. that compile-to-shell language I've forgotten the name of.
I didn’t miss his point. Back in the 70s, many of the new languages were just a subtle variation on the previous one. It’s only later they evolved into something distinctive.
Which is why I said we’d still be using FORTRAN.
Languages that start out radically different don’t tend to gain momentum. Whereas languages that are familiar tend to grow and introduce new ideas.
Nothing is invented in a vacuum.
Also I completely disagree that one shouldn’t create a hobby project need to be innovative. Sometimes people do create things just because they can. And it’s a good thing too because otherwise we wouldn’t have half the open source software available to us today. Many of which was originally intended for personal use, including Linux.
The problem these days is we’re so brainwashed by stories of unicorn start ups pumped with VC money that now everyone thinks every hobby needs to has a viable business plan underneath. It’s like people have forgotten how to play for fun.
So people should go out and create new programming languages. The worst that would happen is they learn to be a better programmer in their day to day language.
I started making a language, and I took many shortcuts.
I just parse my language, translate it to C, and use C compiler errors.
I don't add new semantics, I just add many things like strings, map, etc to make it usable and fast.
I don't know if it's a good idea and how difficult this will be.
"I don't know if it's a good idea and how difficult this will be."
It is a great idea, if you want to learn about languages!
(So if money is your goal, you may want to reconsider)
Nothing wrong with that - some others do it too. You can even use TCC to do quick test builds and only use Clang/GCC for release builds.
> Java and C#, for being enterprisey
I believe there are far more interesting stuff to learn about these languages, like the whole category of runtimes could have been mentioned, which can directly affect the language design itself (e.g. having GC vs some language feature for managing memory, open vs closed world model, having an async feature in the language or let the runtime handle it, etc)
I’m waiting for a llm focused language. We’re already seeing AI is better with strongly typed languages. If we think about how an agent can ensure correctness as instructed by a human, as the priority, things could get interesting. Question is, will humans actually be able to make sense of it? Do we need to?
How could an LLM learn a programming language sufficiently well unless there is already a large corpus of human-written examples of that language?
LLM could generate such a corpus, right? With feedback mechanisms such as side by side tests.
So… llm learns from a corpus it has created?
It’s basically called “reinforced learning” and it’s a common technique for machine learning.
You provide a goal as a big reward (eg test passing), and smaller rewards for any particular behaviours you want to encourage, and then leave the machine to figure out the best way to achieve those rewards through trial and error.
After a few million attempts, you generally either have a decent result, or more data around additional weights you need to apply before reiterating on the training.
How do you define the goal? This kind of de novo neural program synthesis is a very hard problem.
Defining the goal is the easy part: as I said in my OP, the goal is unit tests passing.
It’s the other weights that are harder. You might want execution speed to be one metric. But how do you add weights to prevent cheating (eg hardcoding the results)? Or use of anti-patterns like global variables? (For example. Though one could argue that scoped variables aren’t something an AI-first language would need)
This is where the human feedback part comes into play.
It’s definitely not an easy problem. But it’s still more pragmatic than having a human curate the corpus. Particularly considering the end goal (no pun intended) is having an AI-first programming language.
I should close off by saying that I’m very skeptical that there’s any real value in an AI-first PL. so all of this is just a thought experiment rather than something I’d advocate.
1. Choose set of code challenges (generate them, leetcode, AOC etc.)
2. LLM generates python solution and seperate python test (as in python test calls code as black box process so it can test non python code)
3. Agent using skills etc. tries to write new language let's call it Shark.
4. Run Shark code against test. If fails use agentic flows to correct until test passes.
5. Now have list of challenges, working code (maybe not beautiful) for training.
A bit of human spot checking may not go amiss!
Yes. The learning comes from running tests on the program and ensuring they pass. So running as an agent. Tests and compiler give hard feedback- thats the data outside the model that it learns from.
I think modern RLHF schemes have models that train LLMs. LLMs teaching each other isn't new.
My knowledge is limited, just based on a read of https://huyenchip.com/2023/05/02/rlhf.html though.
RLHF
Interesting page. The latest language I designed is an stack based intermediate language for a C compiler. Not realy intended for human usage, but readable in the sense that you can compare it with the original C code.
Ha! I just wrote stack based RISC cpu architecture with assembler and now thinking about implementing my own FORTH like lang (niche stack based programming language) compiler.
Fun
What do these languages compile to? What's the build pipeline and runtime context?
This talks about programming language.
Right question is to design own linguistic language common between computer and across human.
Marain
I had some thoughts about designing a new language. However it's a huge undertaking and I don't know the answers to some questions:
1. Is there a need for the programming language?
2.If the answer to the previous question is yes, can I find enough people to help and enough resources?
3. If the answer to the previous question is yes, can we release a MVPin a reasonable amount of time?
4. If the answer to the previous question is yes, what is the chance it will gather a reasonable amount of users?
There are literally tons of programming languages that didn't make it. I wouldn't want to waste my and other people resources.
I made a language for using in another project, so I'll answer your questions
https://www.npmjs.com/package/wang-lang
- this new language looks and behaves exactly like javascript, except it doesnt have "eval" and "new Function", so it is CSP safe. That's the only difference. I wanted to execute dynamically generated code in chrome extension
- llm did most of the work of creating a nearley grammar and associated interpreter (whole thing is bundled, nearley is not a final dependency), elaborate tests make this quite sane to handle
- took me about total of 1 weeks for the initial mvp to try out, and then have been fixing bugs and inconsistencies with javascript behavior, about 1 day a month of effort
- mostly 0
The only reason to create was I couldnt find something similar and it was low effort thanks to llm
I also created another even smaller DSL you can say
https://www.npmjs.com/package/free-text-json-parser
It parses json embedded in plain text
Nice. I built something basically just like this for work for the same reason last year. It only look a few hours though, cause I just used Acorn [0] to parse my JS, then directly evaluated the AST. It also had an iteration limit and other configurable limits so I can eval stuff in the browser without crashing the tab. I did not use an LLM.
[0]: https://github.com/acornjs/acorn
This is exactly what I wanted and couldn't find. Ended up creating along with an interpreter (so slightly easier then walk and execute)
> safe
> llm did most of the work
> it was low effort
I really wouldn’t trust its supposed safety.
csp safe has a particular meaning associated it with. its not a "safe" language whatever that is. chrome webstore team is okay with it and serves my purpose. if you have submitted extensions to google chrome then you would know that any sign of "eval" or new Function in the code will lead to rejection.
So what you want is a linter, not a language
I want to execute dynamically generated javascript looking code in chrome extension without using eval or new function. basically eval without actually using eval.
linter would help me find and avoid usages of eval.
"Just for fun" is always a valid default option. Though many authors don't stop there ;-)
I think most popular languages were started as an experiment in some feature, or to solve a specific problem someone had. Those are good reasons to make a language. I see no reason to make a language just to take attention away from other existing languages. Instead, make a language so you can understand how to make languages. It is 100% doable by one person. It's fun and educational.
Sometimes it might just be a fun project to push yourself. Maybe such a complex undertaking can't be fun indeed lol
My idea of for fun is to release something people will use. I have more fun if I work on something useful. For me is less the journey than the end goal.
I love working on software, architecture, design but only if I see some use.
Of course, for other people, the journey is more interesting than the destination and they have fun hacking stuff just for the sake of it. They discover things and learn new stuff they wouldn't have learned otherwise. And this is a path at least as valid as the other.
I'll try to answer your questions best I can.
1. Yes, as long as there are new machines that need programming, new programming languages will be needed. Today's top languages were built for the machines of the 1970, 80s, and 90s. Tomorrow's languages will be built for machines of today and tomorrow. As Alan Kay put it, if you want to invent a new language, first invent the machine of the future and then build a language for it.
2. No, you cannot. First of all, PL devs are cats, it's very difficult collecting them without financial compensation. So if your plan is to post a language and hope that people will come help you, you'll likely be disappointed. The problem is that everyone else interested in building PLs has their own itch to scratch, and they're not going to scratch yours without some compensation.
You might think "Well I can just raise money to do this", and you would be wrong. First, it's very hard to raise money for PLs. Usually you have to have come sort of cred to do it. I know of only 3 projects to have raised VC money for a PL project, and they each had some success before they had done so: Chris Granger (Light Table), Paul Biggar (CircleCI), and Chris Lattner (Swift/LLVM). Granger's project Eve raised $2M and ran out of money after 3 years; Biggar's project Dark also raised money, then fired all the devs when he realized he was burning cash too fast, then he slow-burned development for years, then he gave up and handed development over to someone else; and Lattner raised almost $100M for Mojo, which is probably going to end much the same way as Eve and Dark, but I wish them the best.
Anyway, the point is that you personally (no offense) don't have the profile to raise $100M like Lattner. $2M is not enough for a PL project. Lattner is keeping Mojo closed source for now because there's no good answer for how they're going to make enough money as an open source language to justify raising $100M.
And the reason it's so hard to raise money is because there's no money to be made. No one pays for PLs. No one pays for PL dev tools. They have to be open source or they're rejected by the dev community. The only ones these days who can reasonably pay for all of this with no potential revenue stream are giant corporations, who use the lang as a hook into their ecosystem.
3. Even though the answer is no, you yourself can still get an MVP off the ground in a pretty reasonable amount of time. It's never been easier to make a PL. The problem with PLs is building them is kind of like measuring the coastline; language projects are fractals -- there's an infinite amount of detail you can work on in any given direction. It's very easy for a language project to become a language + editor project, and it's easy for that to turn into language + editor + operating system if you're not disciplined. Plenty of PL devs have fallen into that trap.
4. Rounds to 0% chance. You'll be lucky if you build something that even you will use. Rather, most PL devs end up working on their language in some other language, because working on languages is what they want to do!
That said, it's still important to write languages that you understand no one will use. First it allows you to try new things that may good but unpopular. If PL devs only did what was popular with devs, PLs would go nowhere as a field.
Consider the so called "Hornet's nest" of programming languages [1], which is the tightly related cluster of imperative programming languages which have been the most researched and used over the last 50 years. There is a vaaaaaaaast design space outside that nest, begging for more language development. No one will use most of them, but it's important to understand what those languages might look like to maybe find some new ideas that work.
Also "didn't make it" is kind of an unfair judgement. Gaining popularity doesn't have to be a goal. In fact, it shouldn't be a goal if you want to have any fun at all. There's an infinite amount of work to be done, and if you're not doing it for you, you won't get far at all. That's really the only way to fail at this.
Good luck!
[1] https://tomasp.net/techdims/#footer=index,navigation;left=ca...
Bold plus, making PLs is a lifestyle, not a business. Most PLs clones each other and absorb features. The only difference is QOL and tooling. Users expect to have a full set of batteries, an IDE/LSP, jobs, OOP style, and minimal effort to learn. Being popular contradicts with the idea of pushing the boundaries and shifting paradigms.
This is bad and reads like AI slop. Try the "programming language checklist" instead.
I can't comment on the quality (I don't know anything about PL design) but the page is older than LLMs.
http://web.archive.org/web/20170506182606/https://cs.lmu.edu...
I have to say that the 2017 version is a lot less intimidating though.
N=1: Skimmed through it and found nothing screaming AI.
I think at this point we have to assume all articles are attempts to make some llm better, so they're spewing this shit out and if it gets a genuine reaction, we just gave the llm a rf-cookie. And if someone says "crap llm post" then it gets punished.
It's worth for this SnekQL mascot alone: https://github.com/jennashea/snekql/blob/master/docs/images/...