I've said for a long time that composability in software is a bit like playing Tetris: the lines have to clear.
I feel like that gives an even more literal tower-rising metaphor, and that's what it feels like people using agents naively (and software engineers of lower skill or earlier-career), end up violating.
Agents are getting better at folding things into themselves, especially if you direct them to... but unfortunately I've found that the architectural instincts, even of Fable and 5.6 Sol, are still wildly behind what I reflexively achieve, say.
For sure there is an ability to have agents go back over work and try to fold it into better and better abstractions until it's sort of annealed into something good. I've done something similar on codebases that I have, but the 'high reaches' of architecture with great _prediction of how the software will evolve in the future_ in _subtle_ ways – those are, for now, out of reach of agents.
There is a part of me that wonders if it's partly just how much they can hold in their head right now, though. Even with the greatest articulation and high density of feeding them, the current setups don't allow them to hold a high-quality, sparse, 'zoomable' model of the world in their head that well yet, which we can do pretty well.
But the fact that I'm talking about it in terms of that kind of subtlety is itself promising, I guess?
The upper bound on program complexity used to be the power of the human mind.
"Vibe coding" can break through that barrier. But not because the problem being solved needs that complexity. Because the process does not drive itself towards compact abstractions. It's the AI-powered version of the scaling problem Brooks described back in "The Mythical Man-Month". The combinatoric problems get worse with scale. Concretely, multiple similar implementations of roughly the same thing appear in different parts of the project. This is a known problem of vibe coding now.
We need some way to make AI-driven coding strive for parsimony.
Why would it? It has optimized what it was built to optimize: this is the token-selling industry. Take note that the people hawking the dream of a gold rush are not actually mining but selling shovels
Same issue happens in models trained by organizations who aren’t selling tokens. I believe it’s because being parsimonious is simply harder. Achieving the task at hand independently and declaring the job done is easier than building an abstraction and reconciling between every use case.
Unless we are planning to deploy them all individually to an expensive serverless platform like Lambda, the coordination challenges and overprovisioning are going to more than outweigh whatever architectural benefit you reap (in human-centred development, micro services are solving an entirely different problem - Conway's Law)
Agreed, and ever since LLMs started being able to write competent code, I've noticed a massive difference in quality on codebases where I knew the technology, and ones I didn't. This is because I can much more efficiently steer the LLM on e.g. backend code, which is my expertise, vs yoloing everything on mobile, where I have no idea.
The codebases using technologies I have no idea about tend to quickly become unmaintainable and buggy, because the LLM still doesn't make good architectural choices, but the codebases that use technologies I'm familiar with basically never devolve into unmaintainability.
The difference between the two is massive, and that's why I think that a competent engineer steering an LLM in their area of expertise gets two orders of magnitude more productive, whereas someone steering an LLM in an area they know nothing about are basically producing tech debt at the speed of thought.
AI replaces a single tower with millions of 5-over-1s[1]. The aggregate height, and speed of construction mind-boggling, but when each building is considered individually, not very impressive.
1. Perhaps with a handful of skyscrapers sprinkled in.
The core thesis of this essay is reminiscent of the Lisp Curse [1] / Bipolar Lisp Programmer [2].
It's been a few years since I read these, but if I recall the argument there, it was that Lisp makes it so easy to build stuff and scratch exactly your own itch, that there's no real strong push for lisp programmers to come together and collaborate to build non-trivial and general purpose artifacts. And that is why the landscape of public lisp software is poorer as a result, compared to languages which demand much more effort to get anything substantial done.
Armin seems to be making a very similar point about AI coding.
> There is the appealing idea that AI-assisted programming means better tools which lets us build more ambitious software. That is certainly true at the level of the individual and without doubt a developer with an agent will be dramatically more capable of changing a codebase. But large software projects have never been limited only by how quickly an individual can produce code. They are limited by how well people can coordinate their understanding of the system they are changing.
So true.
Since Nov 30, 2022 everything has become… more complex.
I feel like with software, things have gotten way too complicated (just layer's upon layers upon layers). But to deal with that complexity, now we're using something that just creates WAY more complexity. I've been coding for a while, and I remember the 90s and early 00s where people could make pretty powerful applications with like visual basic or php with essentially no formal training. Those technologies weren't great, but they were really simple and easy to pick up. In contrast, if you try to pick up web development or desktop app development today, it's absolutely overwhelming. Like, something like React is useful but the amount of things you need to know to use it properly is pretty high.
I think introducing AI to deal with this is overall a mistake though. We're just adding more complexity on top of the existing complexity. At best, it's a massive waste of hardware. At worst, we'll probably have agents introducing as many bugs as they fix as they also drown in complexity, and a lot of stuff built using these techniques are going to be fragile garbage while the overall skillset of humanity diminishes because people aren't learning the skills anymore.
Fundamentally, software does not need to be this complicated and it's a solvable problem, but it does require people that care about craftsmanship.
And what's ironic is that a lot of those layers and complexity were added with the stated goal of making it easier for average developers to build applications.
> They are limited by how well people can coordinate their understanding of the system they are changing.
It's not really news, though. Programming as Theory Building (Peter Naur) was published in the 80s, I think?
Maybe the younger entrants to this field never came across it, but even if you never came across it, it was common knowledge amongst experienced devs that understanding of the system you are about to change is crucial.
The complexity of coordinating a project involving more than one entity is, of course, an issue across all industrial sectors—just look at the construction industry.
Thanks for mentioning Peter Naur’s Programming as Theory Building (1985).
I would add Fred Brooks and his The Mythical Man-Month.
I don't know. some stuff has gotten less. Major databases now ship effective HA tooling, microservices seem on their way out, structured databases seem to be back in instead of NoSQL.
HTML and pre-rendering are back in, HTMx, liveview
The degaussing of CSS and the hacks we did, hell i was trying to explain how we debugged web pages in IE6 to a younger staff member today.
Some things are more complex, some things got good enough to make them less complex.
>Since Nov 30, 2022 BC everything has become… more complex.
FTFY
Increasing complexity is the story of mankind. It's the story of civilization.
Someone from 20,000 BC would wander around the earth trying to find food, trying not to freeze, and trying not to get eaten. Someone from 5,000 BC would be trying to grow food, hoping it rains, and hoping disease didn't wipe out the village. The second one increases the complexity from all the systems required to manage people and keep the land growing. Today the vast majority of people on earth don't grow their own food at all, and instead are busy in some way managing the complexity of a large society.
Someone from 1970-80 would think our software from pre-llm days was vastly more complex. They'd just code directly to the hardware with no abstraction layer. Now almost no one does that. We abstracted the hardware away in most cases. With cryptography libraries for the vast majority of people it's complexity is abstracted away and mostly people are told "don't try to write your own crypto because you will fuck it up".
The question now becomes, how quickly will LLMs be able to coordinate their understanding of the system they are changing?
Then I think you should check in with your favorite mental health provider before you become a danger to yourself or others.
Simply put LLMs do understand some things within their crystalized intelligence. Your anthropocentric mind may not accept this, but one day it will. As LLMs have a very short context window in relation to their stored knowledge they have very limited plastic intelligence to change their minds or adapt. All of which is flushed away at the end of a session. It would be like living without the ability to turn your short term memory into new long term memories.
I would gladly use another word for what LLMs can do, but the world at large has not adopted any. The definitions we use around intelligence, comprehension, understanding, consciousness, and sapitence have already been failing us for some time before LLMs as our scientific understanding of biology has increased over the decades as it is. I am one for more exacting definitions when they exist, but humans seem to barely understand the inner workings of our own minds, in large such words escape us.
I'll meet you in the middle: an LLM "understands" words in the same way a toddler understands the phrases they say. "My want cookie!" The toddler has zero comprehension of what any of those words mean, but they know that saying them in that order might result in something desirable.
An LLM has zero understanding of "my", "want", or "cookie" because an LLM has no id/ego, has never felt desire, and has never eaten a cookie.
> The shared language of a software project is not English or Python but it is the common understanding of what its concepts mean, where the boundaries are, which invariants matter, who owns what, and why the system has the shape it does. This language is rarely written down in one place. It lives partly in documentation and code, but also in code review, conversations, arguments, and the experience of having to explain a change to somebody else.
This is so true. I am a big fan of Christopher Alexander’s “Pattern Language” concept, which addresses this exact problem! In fact he recommends developing your own pattern languages for your own domains (which of course led to the famous GoF Design Patterns book).
I have been experimenting with a “Pattern Language” skill which instructs the AI to maintain 3 pattern languages for every project. One in the business domain, one in the product domain, and one in the technical domain. It is working really well. It is always super cool to see it reference the pattern languages during planning and curate them during implementation and review.
I credit using it with keeping my 100% ai-coded projects well organized, aligned across domains, and easy to work on.
My comment is not directly responding to the essay, but it got me thinking about about how agentic programming is much more akin to management than it is to actual programming. Managers generally only have a high level idea of what ICs are working on and often don't have the time, bandwidth, and in some cases ability to understand everything the ICs they're supervising are doing. As more and more software gets written agentically the role of software engineer becomes less technical and more managerial.
It feels to me like I'm stuck doing code reviews for a junior dev all day so I use it as little as possible and mostly to look for things I may have missed.
For example, yesterday I came across some unit tests that didn't have error messages in their assertions. Normally, it takes me ~10 minutes to fix a handful of tests in this situation. In this case, I gave a 2-3 sentence prompt, went to the bathroom, and reviewed the result after I washed my hands. Saved me a bunch of time!
I encourage you to accept a feeling of "imposter syndrome" when using it, and keep trying new things with it. Don't feel like you have to be hands off, except when you're confident that you can be. (IE, if you think you need to spend 30+ minutes on mindless refactoring, see if you can explain it to an agent and then look at HN while it runs. You might get a good result, otherwise, it probably was time for a break anyway.)
BTW: It's important to try different models. The Claude 5.0 models are slow and give me bad results, so I'm sticking with 4.x for now.
I come back to Babel and the Bruegel image too, although taking from it a little less optimism.
I feel these systems rising and sprawling with wee myopic agents developing out their little corners of this unknowably vast whole… a tower with 50 parapets on one side and some wacky cantilevered maiden tower on the other, and a very serviceable adobe roof over some patio for god-knows-why, and thatch over the landing next to it…
Some grotesque fatberg of designs that make sense at the level of individual design efforts, but that lack the fractal sort of levels of policy and judgment that unify the overall enterprise.
The overall language, as it were.
And language takes discipline to establish and maintain through any sufficiently large group of people—witness the company-speak or army-speak of pretty much any successful organization.
We feel like we’ve conquered the problem of talking the same language as our “Gastown Mayors” (who in turn are talking the same language as their “polecats” and so on all the way down the chain of golems)… but it’s only when it’s all built that the good Lord will humble us… that we’ll realize the understanding we thought we’d transmitted perfectly from our thrones wasn’t quite so shared as we’d imagined.
Three or so years ago, Omar, the creator of DSpy pointed out on Twitter that ~LLMs get better most by better internal collaboration. Wish I could find it.
It seems to me that LLMs and particularly chatbots have already allowed for bigger scale collaboration within the LLM companies versus what was possible within the prior cohort of big platform companies.
Has the result just been taller towers, or actually a change of what is possible?
It used to be that you need a good reason to make huge refactorings, because it’s often so much work. Now agent can rewrite half of your code if your prompt is vague enough and you don’t actual try to review it all. And so the “soul” of a program can change dramatically every single day. It’s both great and very much not so.
The biggest obstacle to huge refactoring has always been minimizing the risk of bugs, not losing any features, and ensuring compatibility with the existing ecosystem. The reason it's become easier in the age of AI is because we stopped caring about these things.
Yep. That’s what people are forgetting. If you have an application that many people depend on to do real work, to make money, you won’t survive if you allow AI to constantly make huge changes.
Your test suite doesn’t cover all workflows. It doesn’t cover every combination of actions a user can take. So every big AI refactor while change some of those.
If this is happening frequently, your software will feel like a janky piece of unusable crap.
I don't know why people hold on to all this extra software and features when with the tools its easier than ever to strip that out and refactor the end product in to a much more compact deliverable. Maybe once upon a time it was useful to keep legacy parts of the software solution around, but it can be recreated with fresh eyes if needed given the power of the new LLM models. My philosophy is if its not needed, it needs to be removed.
I interpret "keeps rising" negatively. Changes keep getting made, certainly. The AIs will perhaps never fail to fulfill your feature request. But there's no overall plan. It's just undirected, cancerous growth. It's Homer Simpson telling a team of automotive engineers to add feature after feature.
This isn't really a good way to judge things. In the future, the fondest memories someone else has about technology will be about the present. The past is not better, you're just nostalgic for it.
Every time I think the past was better, I think about how terrible ksh scripting was in 1995. And look at how great peoples' bash scripting is now compared to when we though bash had reached its apex in like 2009.
Is growth enough if technology makes our lives worse? Is a tower the pride of the civilization if a strong gust of wind could bring it down? It is before the gust, when all that matters is that the tower is tall rather than strong. After the gust, things are a bit more nuanced. Fingers are pointed.
The tricky part here is that you can't tell if a once-topmost part of the tower is sturdy until a great deal more tower is resting on it. Well, now a lot the economy is resting on little other than AI dreams. Your move, rational people.
The agent will always fill in the gaps in your understanding. It's not a compiler. It's categorically different from any of the other ways we've built software.
I'm not sure reading code is coming back. The ritual of reading code must come back, because that's the only way to build products that don't collapse under their own incoherence, both technically and visibly.
"just ask Claude" is fine, but it's not the end state
You use a shared agents.md and an auto updated architecture doc but that is the one that needs to be heavily scrutinized and everyone gets a turn to review it.
this doesn't work in any truly complex system. If the entire organization's shared understanding could be captured in a few documents, software engineering would've been a solved problem ages ago.
I feel like that gives an even more literal tower-rising metaphor, and that's what it feels like people using agents naively (and software engineers of lower skill or earlier-career), end up violating.
Agents are getting better at folding things into themselves, especially if you direct them to... but unfortunately I've found that the architectural instincts, even of Fable and 5.6 Sol, are still wildly behind what I reflexively achieve, say.
For sure there is an ability to have agents go back over work and try to fold it into better and better abstractions until it's sort of annealed into something good. I've done something similar on codebases that I have, but the 'high reaches' of architecture with great _prediction of how the software will evolve in the future_ in _subtle_ ways – those are, for now, out of reach of agents.
There is a part of me that wonders if it's partly just how much they can hold in their head right now, though. Even with the greatest articulation and high density of feeding them, the current setups don't allow them to hold a high-quality, sparse, 'zoomable' model of the world in their head that well yet, which we can do pretty well.
But the fact that I'm talking about it in terms of that kind of subtlety is itself promising, I guess?
We need some way to make AI-driven coding strive for parsimony.
The codebases using technologies I have no idea about tend to quickly become unmaintainable and buggy, because the LLM still doesn't make good architectural choices, but the codebases that use technologies I'm familiar with basically never devolve into unmaintainability.
The difference between the two is massive, and that's why I think that a competent engineer steering an LLM in their area of expertise gets two orders of magnitude more productive, whereas someone steering an LLM in an area they know nothing about are basically producing tech debt at the speed of thought.
1. Perhaps with a handful of skyscrapers sprinkled in.
It's been a few years since I read these, but if I recall the argument there, it was that Lisp makes it so easy to build stuff and scratch exactly your own itch, that there's no real strong push for lisp programmers to come together and collaborate to build non-trivial and general purpose artifacts. And that is why the landscape of public lisp software is poorer as a result, compared to languages which demand much more effort to get anything substantial done.
Armin seems to be making a very similar point about AI coding.
[1] https://www.winestockwebdesign.com/Essays/Lisp_Curse.html
[2] https://www.marktarver.com/bipolar.html
So true.
Since Nov 30, 2022 everything has become… more complex.
I think introducing AI to deal with this is overall a mistake though. We're just adding more complexity on top of the existing complexity. At best, it's a massive waste of hardware. At worst, we'll probably have agents introducing as many bugs as they fix as they also drown in complexity, and a lot of stuff built using these techniques are going to be fragile garbage while the overall skillset of humanity diminishes because people aren't learning the skills anymore.
Fundamentally, software does not need to be this complicated and it's a solvable problem, but it does require people that care about craftsmanship.
It's not really news, though. Programming as Theory Building (Peter Naur) was published in the 80s, I think?
Maybe the younger entrants to this field never came across it, but even if you never came across it, it was common knowledge amongst experienced devs that understanding of the system you are about to change is crucial.
Thanks for mentioning Peter Naur’s Programming as Theory Building (1985).
I would add Fred Brooks and his The Mythical Man-Month.
The news is that Agentic Programming has made this always challenging task even more challenging.
HTML and pre-rendering are back in, HTMx, liveview
The degaussing of CSS and the hacks we did, hell i was trying to explain how we debugged web pages in IE6 to a younger staff member today.
Some things are more complex, some things got good enough to make them less complex.
Which ones? PostgreSQL doesn't have HA in core.
FTFY
Increasing complexity is the story of mankind. It's the story of civilization.
Someone from 20,000 BC would wander around the earth trying to find food, trying not to freeze, and trying not to get eaten. Someone from 5,000 BC would be trying to grow food, hoping it rains, and hoping disease didn't wipe out the village. The second one increases the complexity from all the systems required to manage people and keep the land growing. Today the vast majority of people on earth don't grow their own food at all, and instead are busy in some way managing the complexity of a large society.
Someone from 1970-80 would think our software from pre-llm days was vastly more complex. They'd just code directly to the hardware with no abstraction layer. Now almost no one does that. We abstracted the hardware away in most cases. With cryptography libraries for the vast majority of people it's complexity is abstracted away and mostly people are told "don't try to write your own crypto because you will fuck it up".
The question now becomes, how quickly will LLMs be able to coordinate their understanding of the system they are changing?
I think the next time I see "LLMs" and "Understanding" in the same sentence, I am going to lose it....
Then I think you should check in with your favorite mental health provider before you become a danger to yourself or others.
Simply put LLMs do understand some things within their crystalized intelligence. Your anthropocentric mind may not accept this, but one day it will. As LLMs have a very short context window in relation to their stored knowledge they have very limited plastic intelligence to change their minds or adapt. All of which is flushed away at the end of a session. It would be like living without the ability to turn your short term memory into new long term memories.
I would gladly use another word for what LLMs can do, but the world at large has not adopted any. The definitions we use around intelligence, comprehension, understanding, consciousness, and sapitence have already been failing us for some time before LLMs as our scientific understanding of biology has increased over the decades as it is. I am one for more exacting definitions when they exist, but humans seem to barely understand the inner workings of our own minds, in large such words escape us.
An LLM has zero understanding of "my", "want", or "cookie" because an LLM has no id/ego, has never felt desire, and has never eaten a cookie.
https://en.wikipedia.org/wiki/VM_(operating_system)
This is so true. I am a big fan of Christopher Alexander’s “Pattern Language” concept, which addresses this exact problem! In fact he recommends developing your own pattern languages for your own domains (which of course led to the famous GoF Design Patterns book).
I have been experimenting with a “Pattern Language” skill which instructs the AI to maintain 3 pattern languages for every project. One in the business domain, one in the product domain, and one in the technical domain. It is working really well. It is always super cool to see it reference the pattern languages during planning and curate them during implementation and review.
I credit using it with keeping my 100% ai-coded projects well organized, aligned across domains, and easy to work on.
Padmé: "For the better, right?"
Anakin: (gazes in silence)
Padmé: "For the better, right?"
For example, yesterday I came across some unit tests that didn't have error messages in their assertions. Normally, it takes me ~10 minutes to fix a handful of tests in this situation. In this case, I gave a 2-3 sentence prompt, went to the bathroom, and reviewed the result after I washed my hands. Saved me a bunch of time!
I encourage you to accept a feeling of "imposter syndrome" when using it, and keep trying new things with it. Don't feel like you have to be hands off, except when you're confident that you can be. (IE, if you think you need to spend 30+ minutes on mindless refactoring, see if you can explain it to an agent and then look at HN while it runs. You might get a good result, otherwise, it probably was time for a break anyway.)
BTW: It's important to try different models. The Claude 5.0 models are slow and give me bad results, so I'm sticking with 4.x for now.
I feel these systems rising and sprawling with wee myopic agents developing out their little corners of this unknowably vast whole… a tower with 50 parapets on one side and some wacky cantilevered maiden tower on the other, and a very serviceable adobe roof over some patio for god-knows-why, and thatch over the landing next to it…
Some grotesque fatberg of designs that make sense at the level of individual design efforts, but that lack the fractal sort of levels of policy and judgment that unify the overall enterprise.
The overall language, as it were.
And language takes discipline to establish and maintain through any sufficiently large group of people—witness the company-speak or army-speak of pretty much any successful organization.
We feel like we’ve conquered the problem of talking the same language as our “Gastown Mayors” (who in turn are talking the same language as their “polecats” and so on all the way down the chain of golems)… but it’s only when it’s all built that the good Lord will humble us… that we’ll realize the understanding we thought we’d transmitted perfectly from our thrones wasn’t quite so shared as we’d imagined.
It seems to me that LLMs and particularly chatbots have already allowed for bigger scale collaboration within the LLM companies versus what was possible within the prior cohort of big platform companies.
Has the result just been taller towers, or actually a change of what is possible?
Why being one (I see as collaborative) was it not desired? Interpretations? Why is it seemed *more* harmful rather than good?
Your test suite doesn’t cover all workflows. It doesn’t cover every combination of actions a user can take. So every big AI refactor while change some of those.
If this is happening frequently, your software will feel like a janky piece of unusable crap.
Is that because of the technology or because of who you were at the time?
The tricky part here is that you can't tell if a once-topmost part of the tower is sturdy until a great deal more tower is resting on it. Well, now a lot the economy is resting on little other than AI dreams. Your move, rational people.
I'm not sure reading code is coming back. The ritual of reading code must come back, because that's the only way to build products that don't collapse under their own incoherence, both technically and visibly.
"just ask Claude" is fine, but it's not the end state
"we can, so we should".
It ended badly.
Where the "tower" was once a company (or team?) of human devs, it can now be a single dev and their agents.
The right engineer can likely replace non-technical co-founders with a couple LLMs. Geez, I can't wait to write that article...