If you want to dive into Lisp, read Paul Graham’s book “ANSI Common Lisp”. It will show you the “why” and the concepts. Then learn Clojure for the syntax.
Even if you can never see yourself programming with Lots of Silly Parentheses (LiSP), the book is worth reading for the concepts. (And to imitate his writing style.)
In the era of LLMs, I often stop to think of Clojure. I believe it had a significant role in popularizing the concept of immutable data structures. Immutability simplifies the required context for understanding and solving problems. This benefits both humans and LLMs.
Would an LLM, or a person solely coding using them, have had the insight to work out that immutability is the way to go, and we should revolutionize software develoment? If LLMs had happened 15 years ago, would all code still be "place oriented programming", with higher token counts and less reliability than we have today thanks to better practices we've learned in the recent past?
If so, what are we missing out on in the future, now that stopping and thinking about code is frowned upon?
Clojure and Elixir both do really well in the LLM age. Functional programming in general but those too in particular are they have a runtime/shell that agents can interop with directly to view/modify the code as it runs.
I actually just wrote about doing more clojure a few days ago
>If so, what are we missing out on in the future, now that stopping and thinking about code is frowned upon?
You're probably asking this rhetorically, but if you were seriously asking, this would be one of those riddle sentences like "This sentence is a lie." If somebody had an idea that they could articulate in natural language on this forum, the natural language would be something an LLM could process and turn into better code. Any true answer could make itself false.
What we'd miss out on if stopping and thinking about code was no longer a thing, would be ideas that stem from direct human experience. Language models relate words to words. There's a lot of power in that, way more than one would think, but they don't relate words to direct experience, except as described in other words. Innovation comes when humans express their direct experience in new natural language, that other humans can benefit from, and eventually language models benefit from.
This comes down to the question of how much LLM reasoning differs from human reasoning qualitatively. Maybe for the most part reasoning is just reasoning, so most things that help for human reasoning also help for LLM reasoning, and vice versa.
That's a good question. I do think that for example immutability is easier for LLMs as well, since it reduces the required amount of context for any give part of the software. But there might be even bigger wins out there for LLMs.
Well answering my own question, but I think a strong type system with clear error messages is extremely useful for an LLM.
Just look at how a system like LEAN allows LLMs to incrementally solve a problem. It absolutely wouldn’t get that far if it couldn’t rely on some type of compiler to verify things.
So yeah, a strong type system that allows lots of constraints to be expressed easily.
I think Haskell and Rust are much more “LLM-compatible” than Clojure.
And I say this as someone who has written Clojure full time for over half a decade.
Oh, absolutely. LLMs thrive with problems that are easy to verify. They don't always get it right, but the number of attempts they can do in a short time more than covers for it if verification is fast. Type systems are great for this.
There was a time in university where I started my days with doing some 4clojure puzzles. I became so obsessed, I managed to solve almost all of them. It's wild what an impact these few weeks had on my programming abilities. It's a beautifully designed language that makes you think in data structures and their transformations.
As a side note: I code Clojure in Lightroom. It was so cool to see your code instantly evaluate and trace through your functions.
By a weird stroke of luck, clojure was one of the first languages I learned. And yeah, I guess it left a lasting impact on me in the form of an appreciation for immutability and dataflow programming.
Unfortunately it also left me with an extremely deep distaste for dynamic types, and in particular, the blind usage of lists and maps for everything. The ML family of languages finally gave me a paradigm worth keeping around.
I always thought that a good craftsman knows many tools and techniques, so I learned and practiced everything. My goal was not to leave blank spots on my technology and language map. I learned FP, OOP, and various architectures like DDD, and I did not dismiss any language or major framework. "Every tool has its own place and is useful". Then, after years, I was employed by a company that used Ruby as their main language. "Such beautiful, simple code," I thought. It was a real culture shock for me. Strings and maps everywhere. No types. Monkey patching. No structure. I was really disappointed, but other words come to mind. I quit. So yeah. Once you go to some regions, there is no way back. I wonder how people who started with, for example, Haskell, and were shaped in a correct way, function when they touch other paradigms.
I started learning Clojure when I was 17, but then later moved away from programming. It's done the opposite with me, I struggle to really get the benefit of static types and I still think of everything as maps and functions.
The biggest benefit of static typing is that you automatically know every single operation you can perform on every single result of every single function ever made.
Haha same! All you need are primitives and maps and lists and first class functions! (Though good decomposition syntax is up there as well; it can be really helpful for telling what a given thing needs the map to include!)
> Awesome, how do you ensure clients of your code always do the correct operations on your data?
Either the data type is opaque (as in you shouldn’t mess with it and you should use the provided api) or it’s just normal primitives and thus no restrictions.
You don’t usually reach into data structures to modify their properties, especially if it’s for foreign modules.
> the blind usage of lists and maps for everything
it's not blind, tho it could seem that way, and if someone hadn't thought deeply abut how they should use lists and maps, then it could be described as being blindly using them.
But this style is also what makes clojure dynamic, and flexible. Updates don't have to be rigid, and keeping backwards compatibility is a lot easier than a lot of statically typed languages.
That's not to say you can't have some rigidity where it's needed (specs for example, can be sprinkled in where you need/want it).
The interesting thing about Clojure, for me, is that for all its beauty and power, it’s so simple that there isn’t much variety in the articles you can write or the “gotchas” you can cover for a wide audience. Niche stuff that assumes you already know Clojure doesn’t catch on with a wider audience. Its strong points—“everything is data,” “REPL development,” “immutability and strong primitives,” and “paradigm shift”—get repeated so often that they become noise people just filter out.
There isn’t much room for yet another framework or another way to work with data (I’m looking at you, Node.js), so there isn’t much to talk about or write articles about.
In other words, the efficiency and simplicity of the language act as a barrier to its spread in a world where only repeated mentions get through. IDK, maybe this is for the better after all.
It's an interesting question, even outside programming.
Before people might ask questions and discuss things on obscure forums or subreddits. Now that LLMs are trained on all this data people no longer need to do that.
You just ask and LLM and get a throwaway answer that people generally trust as the best option.
Where do the new insights and opinion come from? Especially when LLMs start getting trained to recommend certain products.
We don't do that anymore. I mean FP is just natural language now, can't wait for LLM optimized languages. I don't read the rust/ts code produced by claude anymore, so why bother using said language.
So, people say this about programming languages all the time.
My problem is:
a) this assumes that people can not have ideas without a programming languages. That's obviously incorrect. I can have ideas without knowing a programming language.
b) languages also constrain the thinking. For instance, in ruby you have a class and a module, as distinction. Is that a useful distinction? I think it is overlapping, with modules giving more flexibility. In practice, this distinction does not lead to many problems, but if you have: class Foo; module Bar; end; end and then you were to write "class Foo::Bar" in another file because you want to modify the code quickly, then ruby says you'd have to use "module Foo::Bar" instead, because the ruby parser already knew that this is a module, not a class. It is not a big problem, but whenever I run into this issue, I wonder why I need to care whether it is a class, or a module. Because, when I do so, such as adding a new method to that, perhaps from another namespace/project, then I wonder why ruby does not prioritize me wanting to change the code, but instead complains about that it is a module or a class. (I understand the ruby parser has to work that way, in order to differentiate between errors made or mixing up code, but my point of contention here is the "class versus module" distinction. The above is one example of many more that can be found in programming languages.
Of course you can reason that better programming languages typically lead to better implementations too. I'd reason in favour of such a statement. I don't really see the point of learning a language that is inferior to languages I already know, when I know I am not going to use that language. The "learn a new language every year" is not a bad idea, but why would I want to learn languages that are bad? That makes no sense. There is competition between languages, as well as against time.
To me, Clojure was "Lisp for JVM". Is that an incorrect picture?
I mean, recently I saw Clojure for . Net and Clojure without JVM (forgot the name). If you don't need/want JVM, then why would you use Clojure over e.g. Common Lisp?
It used to be tied heavily to Java and then JavaScript but it's on quite a few platforms now. But clojures big thing is dynamic functional programming, passing immutable maps to functions.
Common Lisp was designed to unify several Lisp dialects. Its standard reflects historical compromises, but it also deliberately supports several programming styles.
Clojure makes stronger choices about the default style: immutable data, transformations over collections, and explicit tools for managing changing state.
Zooming out you're right: it's lisp and jvm is a difference. Zooming in and there are different ergonomics and incentives on how to structure the program.
I think clojurists overestimate how advanced their language is. Or maybe it's just my outsider's perspective? Famously, Clojurists/Lispers think that theirs is the most advanced language and everyone who fails to see that just doesn't understand expressive power.
In any case: Typescript has dynamic types and static types, it has true union types, it has hot reload (what you might call REPL), it has the widest package ecosystem and so much more. It works out of the box with the most popular IDE and you don't need to learn emacs.
I rarely hit the power ceiling of Typescript, so I never put in the work to learn Clojure.
Typescript doesn't have a real repl. You cannot redefine just one variable in a running program and have the rest of the compilation untouched, yet aware of the redefinition.
I get that "hot reload" can feel similar, but it's not the same and is more limited for debugging, experimentation, extremely fast (sub-ms) iteration loops, etc.
Another thing you can't do is running/debugging tests without triggering more compilation. In a real repl language you drive things in-process, not from the outside with compilers and test runners.
Typescript is lovely btw, just have a go at other paradigms if you feel like it.
confluent persistence, on data. which changes how you think about concurrency.
it allows you to "branch, work on a worktree, and merge" but unlike git worktrees you don't need to copy a complete clone of the data, you instead, like in the git object store, share most of it and only need memory for what you touch in your branch.
and your data only changes when _you_ decide to rebase, and merging partial work results is a _deliberate_ step, not something happening under your feet.
Outdated article post-LLMs, if you are still taking to heart what this article says you are working against the trend.
You don't need any inner working of clojure or any lower language for that matter. It's all available within your reach. Don't let the people who have built an identity over the tools that they have chosen, they can't give you any meaningful advice anymore with what we have as tools.
I'm saying this as someone who has worked with clojure for over a decade. There is no point in gate keeping things anymore. It also means that the point of clojure which is supposed to have a lot of DX built into it , is gone with LLMs.
Software is largely solved now and the only thing that matters now is distribution so start thinking about your product and how to market it.
Outdated take post-slopocalypse. You still very much need to know how to program and will often get better results writing code yourself, as we have seen from the outages caused by people vibe coding shit.
> if you are still taking to heart what this article says you are working against the trend.
Being trendy is not synonymous with being correct. In this case the trend is stupid and making the profession worse, so one would be well advised to work against it.
> Software is largely solved now
If that were true, we would be getting a deluge of high quality software. Instead we are getting a deluge of crap. Software is by no means solved.
While I agree "LLM" have lowered the barrier to access languages, it is fundamentally just isomorphic plagiarism of around $9Tn worth of FOSS projects.
Much like how GPU were bodged onto the monolithic von Neumann CPU architectures, we may safely assume most "LLM" will be superseded by neuromorphic computing technology sooner of later. =3
Viewing LLMS as compression is flawed because it shouldn't be possible to solve new novel problems. Zipping every single file in existence and then asking Winzip to produce a novel working zip file clearly doesn't work. Clearly LLMs are learning from data it is fed and a lot of what we do before LLMs also involved "plagirism". Even the greatest writers and artists would copy each other to master their styles and extract the dimensions. I think framing LLMs as compression is not only silly it demonstrates a poor understanding of the inner workings of LLMs
Except a statistically salient search context is in proximity to something similar, and manual content curators drop their thumb on the scales weights to boost desirable outcomes in the quantized model.
> Much like how GPU were bodged onto the monolithic von Neumann CPU architectures, we may safely assume most "LLM" will be superseded by neuromorphic computing technology sooner of later. =3
You're absolutely right! It's all in the title really, Clojure can't affect how we think about programming anymore, because we stopped both thinking and programming.
If you want to dive into Lisp, read Paul Graham’s book “ANSI Common Lisp”. It will show you the “why” and the concepts. Then learn Clojure for the syntax.
Even if you can never see yourself programming with Lots of Silly Parentheses (LiSP), the book is worth reading for the concepts. (And to imitate his writing style.)
Would an LLM, or a person solely coding using them, have had the insight to work out that immutability is the way to go, and we should revolutionize software develoment? If LLMs had happened 15 years ago, would all code still be "place oriented programming", with higher token counts and less reliability than we have today thanks to better practices we've learned in the recent past?
If so, what are we missing out on in the future, now that stopping and thinking about code is frowned upon?
I actually just wrote about doing more clojure a few days ago
https://tuhat.net/@sbr/p/beam-to-jvm
You're probably asking this rhetorically, but if you were seriously asking, this would be one of those riddle sentences like "This sentence is a lie." If somebody had an idea that they could articulate in natural language on this forum, the natural language would be something an LLM could process and turn into better code. Any true answer could make itself false.
What we'd miss out on if stopping and thinking about code was no longer a thing, would be ideas that stem from direct human experience. Language models relate words to words. There's a lot of power in that, way more than one would think, but they don't relate words to direct experience, except as described in other words. Innovation comes when humans express their direct experience in new natural language, that other humans can benefit from, and eventually language models benefit from.
Phrased differently: what would a language look like that’s optimized for LLMs?
Is anyone out there trying to find these wins?
Just look at how a system like LEAN allows LLMs to incrementally solve a problem. It absolutely wouldn’t get that far if it couldn’t rely on some type of compiler to verify things.
So yeah, a strong type system that allows lots of constraints to be expressed easily.
I think Haskell and Rust are much more “LLM-compatible” than Clojure.
And I say this as someone who has written Clojure full time for over half a decade.
As a side note: I code Clojure in Lightroom. It was so cool to see your code instantly evaluate and trace through your functions.
do you mean lighttable? or you really code in adobe's photo editor.
So you can do this when programming:
[0]: https://github.com/typedclojure/typedclojure/tree/main/examp...[1]: https://github.com/typedclojure/typedclojure
- Clojure for functional programming (you mean you can update a bitmap in a purely functional way??)
- Array langauges (APL, BQN, J) for avoiding loops with matrix operations
- Prolog for avoiding loops with backtracking
MVP churn and AI fatigue is getting to me. Need some good Hammock Time!
Unfortunately it also left me with an extremely deep distaste for dynamic types, and in particular, the blind usage of lists and maps for everything. The ML family of languages finally gave me a paradigm worth keeping around.
With static types it's obvious and you generally have to work hard (and in non intuitive ways) to mess up.
This becomes more important as team sizes grow.
Either the data type is opaque (as in you shouldn’t mess with it and you should use the provided api) or it’s just normal primitives and thus no restrictions.
You don’t usually reach into data structures to modify their properties, especially if it’s for foreign modules.
it's not blind, tho it could seem that way, and if someone hadn't thought deeply abut how they should use lists and maps, then it could be described as being blindly using them.
But this style is also what makes clojure dynamic, and flexible. Updates don't have to be rigid, and keeping backwards compatibility is a lot easier than a lot of statically typed languages.
That's not to say you can't have some rigidity where it's needed (specs for example, can be sprinkled in where you need/want it).
There isn’t much room for yet another framework or another way to work with data (I’m looking at you, Node.js), so there isn’t much to talk about or write articles about.
In other words, the efficiency and simplicity of the language act as a barrier to its spread in a world where only repeated mentions get through. IDK, maybe this is for the better after all.
Before people might ask questions and discuss things on obscure forums or subreddits. Now that LLMs are trained on all this data people no longer need to do that.
You just ask and LLM and get a throwaway answer that people generally trust as the best option.
Where do the new insights and opinion come from? Especially when LLMs start getting trained to recommend certain products.
https://github.com/conductor-oss/clojure-sdk
My problem is:
a) this assumes that people can not have ideas without a programming languages. That's obviously incorrect. I can have ideas without knowing a programming language.
b) languages also constrain the thinking. For instance, in ruby you have a class and a module, as distinction. Is that a useful distinction? I think it is overlapping, with modules giving more flexibility. In practice, this distinction does not lead to many problems, but if you have: class Foo; module Bar; end; end and then you were to write "class Foo::Bar" in another file because you want to modify the code quickly, then ruby says you'd have to use "module Foo::Bar" instead, because the ruby parser already knew that this is a module, not a class. It is not a big problem, but whenever I run into this issue, I wonder why I need to care whether it is a class, or a module. Because, when I do so, such as adding a new method to that, perhaps from another namespace/project, then I wonder why ruby does not prioritize me wanting to change the code, but instead complains about that it is a module or a class. (I understand the ruby parser has to work that way, in order to differentiate between errors made or mixing up code, but my point of contention here is the "class versus module" distinction. The above is one example of many more that can be found in programming languages.
Of course you can reason that better programming languages typically lead to better implementations too. I'd reason in favour of such a statement. I don't really see the point of learning a language that is inferior to languages I already know, when I know I am not going to use that language. The "learn a new language every year" is not a bad idea, but why would I want to learn languages that are bad? That makes no sense. There is competition between languages, as well as against time.
Which is precisely why trying out languages built on different paradigms is a good way to widen the way you think about programming.
I mean, recently I saw Clojure for . Net and Clojure without JVM (forgot the name). If you don't need/want JVM, then why would you use Clojure over e.g. Common Lisp?
Clojure makes stronger choices about the default style: immutable data, transformations over collections, and explicit tools for managing changing state.
Zooming out you're right: it's lisp and jvm is a difference. Zooming in and there are different ergonomics and incentives on how to structure the program.
In any case: Typescript has dynamic types and static types, it has true union types, it has hot reload (what you might call REPL), it has the widest package ecosystem and so much more. It works out of the box with the most popular IDE and you don't need to learn emacs.
I rarely hit the power ceiling of Typescript, so I never put in the work to learn Clojure.
I get that "hot reload" can feel similar, but it's not the same and is more limited for debugging, experimentation, extremely fast (sub-ms) iteration loops, etc.
Another thing you can't do is running/debugging tests without triggering more compilation. In a real repl language you drive things in-process, not from the outside with compilers and test runners.
Typescript is lovely btw, just have a go at other paradigms if you feel like it.
it allows you to "branch, work on a worktree, and merge" but unlike git worktrees you don't need to copy a complete clone of the data, you instead, like in the git object store, share most of it and only need memory for what you touch in your branch.
and your data only changes when _you_ decide to rebase, and merging partial work results is a _deliberate_ step, not something happening under your feet.
You don't need any inner working of clojure or any lower language for that matter. It's all available within your reach. Don't let the people who have built an identity over the tools that they have chosen, they can't give you any meaningful advice anymore with what we have as tools.
I'm saying this as someone who has worked with clojure for over a decade. There is no point in gate keeping things anymore. It also means that the point of clojure which is supposed to have a lot of DX built into it , is gone with LLMs.
Software is largely solved now and the only thing that matters now is distribution so start thinking about your product and how to market it.
What tipped the scales for Clojure - Brendan Foote (Clojure/Conj 2026)
https://youtube.com/watch?v=tzzg8CVkl2A
But they will be overtaken by those who use the LLM as a sparring partner to form their own understanding and elevate themselves.
I fear we will see a huge disparity between both groups. One ever on the rise while the other is left in the dust.
Outdated take post-slopocalypse. You still very much need to know how to program and will often get better results writing code yourself, as we have seen from the outages caused by people vibe coding shit.
> if you are still taking to heart what this article says you are working against the trend.
Being trendy is not synonymous with being correct. In this case the trend is stupid and making the profession worse, so one would be well advised to work against it.
> Software is largely solved now
If that were true, we would be getting a deluge of high quality software. Instead we are getting a deluge of crap. Software is by no means solved.
Much like how GPU were bodged onto the monolithic von Neumann CPU architectures, we may safely assume most "LLM" will be superseded by neuromorphic computing technology sooner of later. =3
https://www.youtube.com/watch?v=LPZh9BOjkQs&list=PLZHQObOWTQ...
The encoding of the LLM is almost always lossy compression, but we agree it is not real intelligence.
As a sentient turnip, I likely wouldn't understand such things, but do recognize copyright bleed-through when it includes known obscure works. =3
https://en.wikipedia.org/wiki/Mechanical_Turk
perhaps this is the optimism i need today.