Can't help but feel that AI is the killing of the golden goose. We thought it would kill jobs, but it may just end up killing entire industries.
Intellectual property incentivized many and laid many golden eggs, but why would someone toil to create a new algorithm to solve a problem if someone else is just going to steal it in hours?
But who knows, maybe imploding the software industry is a good thing. I may lose my house in the process, but I think we could all get back to touching a little more grass.
> but why would someone toil to create a new algorithm to solve a problem if someone else is just going to steal it in hours?
same reason I spent my entire career so far writing open-source code that anyone can copy ? (it's GPL, but imagine 100% companies actually caring about that). Ideas can't be owned.
Intellectual property is terrible for society on the whole.
You could just as easily argue "why are 200 companies burning through people and natural resources to come up with the exact same thing and then marketing it as different" as you can say "why would anyone create anything without the ability to making oodles of money by restricting imitation".
Just stop this nonsense. Intellectual property is good. Free open source is not good. Anything that is free while take a lot of effort is essentially killing potential jobs. Wake up.
In a society rich enough to be able to provide for everyone (which it is) and where food and housing are abundant (which they should be), then things being free isn't bad because you aren't worried about making money to live.
You are the one who needs to wake up from the assumption that the only way to live is to hoard your intellectual products instead of imagining a society where you can stop having to compete on that level.
Not only that, rent seeking destroys jobs as a result of artificial scarcity and hoarding.
You pay $1000 to a monopolist instead of $100 to a small business and then having the other $900 to spend on whatever you want, like patronizing 9 other small businesses and actually getting something more for your money. The $900 in monopoly rent goes to Wall St instead of Main St and is used to e.g. bid up housing costs so people can't afford to buy a house.
The incumbents lobby for restrictive zoning or professional licensing gatekeeping to keep those rents high, which means fewer jobs doing construction.
It doesn't help you for huge corporations to get more of your money.
> But on the other hand, no IP means economy of scale is a lot more important, and that favors huge corporations.
"IP" is the thing that favors economies of scale.
Suppose you're a manufacturing company and you want to make your own computers. You're great at making the chassis and screen and batteries and things like that, but to make a whole device you also need an operating system, a compiler, a web browser, a CPU, etc.
If you're free to use existing ones then you have a chance. If you have to develop every one of those things in-house because every existing one is the property of an incumbent who is using it as a moat then you have to be a megacorp or you're sunk.
Notice that this is what all of the smaller companies actually do. Every small consumer device OEM and some of the big ones are using some variant of Linux/Android or BSD rather than developing their own OS. NeXT, the scrappy company Apple bought and turned into Mac OS X when they were themselves the underdog, was based on BSD. RISC-V has shipped more than 30 billion cores and is taking a decent chunk of the embedded market from ARM, which is itself a company that licenses its designs to companies that compete with each other (and run Linux on their chips) rather than using exclusivity and mandatory vertical integration to try to monopolize the whole supply chain. Which is how the mobile market went to them rather than Intel and Microsoft. Who themselves took the desktop and server markets from even more vertically integrated mainframe companies.
The availability of composition is what allows you to be a small company that does one thing well. Which is the thing you can't do when someone else has a monopoly on something you don't make but need, and correspondingly isn't willing to sell it to someone who is trying to compete with them in a different part of the supply chain.
Moreover, the economies of scale you're talking about on the other side actually have diminishing returns. Suppose anybody who wants to make a product that typically sells for $100 has to buy a $10,000 machine in order to do it, but then can make an unlimited number of them. If you're so small that you only ship 100 units, the cost of that machine by itself is adding $100 to the cost of producing of each unit and you're pretty screwed. But if you're a medium sized company and you ship a million units, it's only adding one cent. The advantage of the medium company over the small one is dramatic. Whereas the megacorp that ships ten billion units shaves it down to 1/10000th cent, but then the advantage of the megacorp over the medium one is still less than one cent.
On top of that, it's often something like you need a $10,000 machine that can make a million units even if you only use it to make a hundred, and then the scale advantage of the medium company over the small one is all still there, but the advantage of the megacorp over the medium one doesn't even exist because then to make ten billion units they need ten thousand machines.
Which is even before we get to diseconomies of scale: The larger an entity is, the more efficiency you lose to bureaucracy, coordination costs, principal-agent problems, internal politics, etc.
There is a reason industries with low barriers to entry tend not to monopolize. Why is there no monopolist who grows all of the coffee or weaves all of the clothing? Shouldn't "economies of scale" allow a single huge company to out-compete all of these little ones in the absence of anyone having a way of excluding competitors?
Conversely, find a consolidated industry and you'll generally find some kind of government-backed monopoly or regulatory capture.
I can't find any sources implying that there is any single company or country that produces the majority of clothing. China and the EU are each around a third with no single country having even a majority of the remaining third, and China and the EU each have multiple internal producers in competition with each other.
Are you objecting to the use of the term "weaving" to describe modern clothing production?
AnthonyMouse gave you a overly lengthy rebuttal without getting to the point but the cost structure of IP is as follows: the marginal cost of making a copy is low enough to be negligible compared to the fixed cost of creating it.
This means the theoretical marginal cost is equal to the development cost divided by the number of paying customers. Where the development cost can either be a one time expense or an ongoing expense.
Companies make the fixed investment (or take someone else's fixed investment like an open source project) and they charge market rate pricing. The difference between the marginal cost (which is dropping with scale) and the market rate is economic rent or producer surplus. Competing companies pop up and work on their own product because they see the producer surplus and want a piece for themselves. This is how you get many companies wast the producer surplus on reinventing the same software over and over again.
Now with open source software you have the opposite problem. The marginal cost is zero, but the fixed cost still has to be paid. This means that the customer gets something that they would otherwise have to pay market rate pricing for free, this is consumer surplus. The marginal cost that the consumer would have to pay to keep the project alive becomes irrelevant the more consumers there are, which makes open source a financial tragedy because it fixed the cost structure problem at the expense of self sacrifice.
The premise is, if you don't have some way of excluding competitors then you're in a competitive market and larger companies have the advantage of lower fixed costs per unit. There are fixed costs other than R&D, like the capital cost of buying manufacturing equipment or regulatory certifications or things like that. In the abstract a larger company can then use that to charge lower prices and out-compete anyone else.
But past a certain size that doesn't actually happen anymore. If you're so small that your costs are 100% higher then you're probably screwed, but if your costs are <1% higher then there are plenty of other ways to make up for that.
> which makes open source a financial tragedy because it fixed the cost structure problem at the expense of self sacrifice.
Except that it actually works fine. You have a piece of community-developed software that it would costs you a billion dollars to produce from scratch and it works except for a few issues that are costing you a million dollars and would take half a million in development resources to fix. Half a million is less than a million and is way less than a billion. So each company hires a small number of developers to make the contributions most relevant to their own business and get what they want, rather than one company hiring a large number of developers to introduce all the principal-agent problems inherent in central planning etc.
This even works where you have 20 companies that are each losing $100,000 for not having a feature that would cost a million dollars to develop, because then instead of one of them contributing a million dollars in resources, they each contribute $50,000. And you don't need there to be perfectly exact accounting here because the company has already hired a couple of developers to add the feature that costs $500k and gets them $1M and can have the same people collaborate with other users on the feature they all need. Or they make a financial contribution along with a feature request and multiple companies wanting the same feature collectively provide enough to develop it.
Where it actually breaks down is where you have companies whose management is too daft to realize that it does work and then doesn't make any contributions. So they lose a million dollars in value because they didn't want to contribute $500k, or they pay a million dollars to license some proprietary software from a company who keeps the other $500k as profit instead of paying half that much to get the existing open source project into the state where they can use it.
Are you from 1995? FLOSS won long ago. Services give companies far more money than software. Closed software it's dead. Custom services and in-place programming (adding features, interop...) for small non-IT companies gives far more money for IT companies.
> but why would someone toil to create a new algorithm to solve a problem if someone else is just going to steal it in hours?
The same reason I have open sourced every line of code I have written in the last several years, even new inventions. Because it is good for society, and good for my reputation to show what I am capable of which has been very good for my career.
Better to be the one to open source it before someone else does.
The only business model left is convenience tax. Save people time and they give you money.
> The only business model left is convenience tax. Save people time and they give you money.
Which is why LLMs are going to kill most of the software industry. Software products and services operate on this assumption - package some functionality into an opinionated UI that makes some kind of work marginally more convenient, slap a brand on top, and you now have a captive audience and the perfect upsale/ad delivery vector, i.e. the user interface.
LLMs flip that around, because suddenly, your UI is an obstacle sitting between the user and the problem they want to solve.
This has been debated about for quite a while, where people give donations in support or the "pay what you can" model.
Unfortunately, many have argued that it (donations or PWYC) doesn't work and people usually pay only when forced to. There is also the claimed psychological issue, where when something is made free or people can pay minimum amounts, that it is devalued or thought as near worthless. On the flip, various people and organizations have claimed that it's possible to use alternate models, depending on how it's done.
Hackers with time on their hands will never pay for anything on a computer they can do themselves, but those with more money than time will pay even if it is open source.
> Because it is good for society, and good for my reputation to show what I am capable of which has been very good for my career.
I feel that LLMs are destroying the reputation building benefits as well. It used to be that most people's published code was a result of their personal effort and thought process. This is no longer the case. Not only are many new contributions written by machine, but the very process of understanding and reasoning about code has been devalued.
Yes, and this is a net positive. The only reputation points that will matter will be building novel things to solve problems no one has ever solved before. Incentives are much better aligned for progress instead of everyone wasting time building yet another todo list app as a resume booster.
Do you think this will still apply for future generations? It seems like many people now are grabbing code either directly from model training or using agents to blindly scrape the web, finding code and customizing it without ever considering the author. Do you think individual contributors of open source code will have the same reputational incentive?
> Do you think this will still apply for future generations?
Not in the same way. Like today if you see a programmer with a bunch of punch cards of assembly they hand coded, that is impressive, but if that is all they have I would question if they can provide value in a modern software org.
The reputation that matters moving forward will be the 100x engineer that is digitally sovereign and built all the software they use from firmware to kernel to UX. The one who automates everything, has a history of making things people want, and understands a system down to the metal. That is the only person I could be convinced to hire at this point, and the person I am striving to become.
it will also help with the environment by stopping a lot of forced obsolescence for no reason (apple rolls out a new OS every year, then magically wacom stops supporting perfectly working tablets in their new driver versions)
Somewhat related, but I sicced an LLM at my wacom's BT and USB packets to connect it to an unsupported system and the dang thing got everything decoded in 15 minutes. Can't wait to resurrect some old hardware I've got laying around.
> why would someone toil to create a new algorithm to solve a problem if someone else is just going to steal it in hours
When that happens it shouldn't be forgotten that people (broadly speaking) chose this. AI proponents lit intellectual property on fire out of self-interest.
The technology is powerful enough that it was bound to happen regardless. If it didn't happen here in the US, it would have happened somewhere that IP laws aren't respected.
I don't think that's the rebuttal you think it is.
GP's argument is usually used in the sense of "if I hadn't done it, someone else would, so I might as well do it and the end result will be the same, except then I'll reap the rewards".
It's a crap justification for doing something unethical. But in many cases, even the unethical ones, it's also true, and I think in this case it is too. If the US had magically shut down every AI lab on day one, and banned all AI research, that wouldn't have stopped this train. Probably would have slowed it down, but not enough to really matter.
I never said it was justified, just observed that it would happen and is happening, regardless of what "AI proponents" want to happen. IP law, like many laws regarding business practices, have and always had varying regulatory environments around the world when countries find it advantageous for their economies to have those environments.
Regardless of whether or not we agree with it, adherence to our idea of IP law is not something the whole world agrees upon, even before LLMs.
> but why would someone toil to create a new algorithm to solve a problem if someone else is just going to steal it in hours?
I think it goes deeper than that, in that the AI era will eventually force us to rethink how we value each other. It used to be economic scarcity: I provide a service that very few people can provide, therefore I make a lot of money (supply and demand.) Now we're moving to a place where most anybody can do anything, so our value system needs to evolve, as well. Instead of valuing scarcity, perhaps we should value personal creativity and self-expression. Kind of like paying money for a concert, or for someone's personal painting, even though you can listen online, or you can just have AI generate the same painting. It comes more from wanting to support people than to supply scarcity-based demand.
I think this idea makes many peoples' eyes roll, but IMO that's a testament to where we are today as a society. Most interactions get reduced to just business transactions without the humanity in it, and that's where our problems come from.
That's a nice vision but if AI and robots end up replacing most jobs then we're first going to suffer a total collapse in living standards across the world, until we replace capitalism at least. When the economy no longer relies on labour people won't have anything to trade in exchange for basic necessities.
The only way AI and robots end up replacing most jobs is if they can do them for less money than people can. Which in turn means that those things become less expensive, since that's the thing causing suppliers to do it.
So there are really three possibilities here. First, we get robots that can e.g. build houses and housing scarcity goes down, so now you don't need to make much money to afford rent. Second, robots still can't build housing, so then you get a job building housing and have a job. Or third, robots still can't build housing but we have a bunch of rent-seeking incumbents that won't allow you to get a job doing that either, because they're trying to keep rents high or artificially limit the supply of trade labor.
It's the third one that really screws you and is the one that we need to prevent.
So now things cost an average of 50% as much. If you were spending $100 before, it's $50 now, and you have the other $50 to spend on something else. Which in turn creates a job doing the something else.
Consider all of the other times things like this have happened. The majority of existing jobs at the time of e.g. the industrial revolution were the occupations that no longer exist or can be done with less than half the original amount of labor. Why then has the modern unemployment rate typically been at <10%? Because then people get other jobs.
The absolute extreme case where all jobs are automated is the one where nobody needs to work because all work is automated and everything is abundant. In every other case, people have jobs doing the things that can't be automated.
If it costs half as much to produce something, is the result that people get more of those things or less? Do more people starve today or in the times before tractors and combine harvesters?
The transition period often sucks, because then you have a lot of people whose skills are misaligned with where the new demand is, but long term that goes away so the real question is what do you want to do about it in the short term.
You jsut start demanding goods and services fall in price by 50% because labor costs have been eliminated or drastically reduced. If vendors demur, then you take stuff or make your own.
We all know that "real" communist societies like North Korea are worse, much, much worse. In fact within China how bad this is varies enormously. If you're in a "rope village" somewhere deep inland, having the finest house in the entire city is arguably worse than homelessness in Beijing, but the party won't let you leave except as homeless, because you're allocated a "Household" (Hukou) and you aren't allowed to leave.
And, of course, half the problem is the very structure of the communist party itself: local officials hide the problem, because if they don't, they'll be purged or at the very least won't see promotions, ever. Which means that the police will be using violence to hide the problem rather than fix it.
AI can provide incredible wealth, or immeasurable misery, in both capitalist and communist societies. That is not what makes the difference, that is just how political parties like to frame themselves and so the argument why you should choose "left" or "right".
Capitalism and private home ownership can certainly provide housing, but if you go too far or not far enough in private ownership and "protecting capital" they won't. And communism can certainly fix homelessness (the Soviets did, for example), except nowadays they just won't.
AI will provide more tools for people with capital, which primarily will be the government. Whether that solves, say, homelessness is a choice, in either system.
you're allocated a "Household" (Hukou) and you aren't allowed to leave.
This is really not accurate. Historically people have been disincentivized to leave by being excluded from social services and so on, the intent being to avoid large cities turn into mega slums. But the Chinese state has begun to dismantle the hukou system in recent years and allow people to move to mid-tier cities without restrictions, and to grant hukou rights based on duration of residency.
You missed the point and went on a tangent about random countries today.
I don't claim to have a solution for what comes after but it is certain that capitalism will collapse if we ever see significant job loss due to AI and robotics.
FWIW, they were right. Society may have ended up better off in the end, but the Luddites - the actual individuals affected by the transition, and their families and their children - were absolutely screwed for life.
1) Capitalism has historically been proven to work with 90% unemployment, unlike communism. Of course that would represent an enormous loss compared to today but it would not be enough to collapse it.
2) The same would happen to communism. Talk to someone from China and ask what their government will do when they can run the economy on 90% less people. The answer I've gotten is that they already know: after all the division between CCP/state jobs (which includes all state companies) and "normal" jobs already works like that. Then it was mentioned that there's a reason that people have committed murders for CCP jobs ...
I don't understand what you're complaining about. Yes, the US and Europe are capitalist, BUT there is more than enough room in the law for communist communities. Some of those existed for decades, and some still exist.
If you truly believe communism is better, why not just join one of those instead of dragging everyone along with you?
> Capitalism has historically been proven to work with 90% unemployment
What an extraordinary claim. Elaborate.
> Of course that would represent an enormous loss compared to today but it would not be enough to collapse it.
Betting against a revolution in such a scenario is bold.
> Talk to someone from China
China is capitalist, not communist.
> If you truly believe communism is better, why not just join one of those instead of dragging everyone along with you
Where did I say that? That's a nonsense argument for various reasons but it's not the point of this discussion so I'll skip it.
The Great depression saw 25% unemployment and that was just a temporary upset, try 50% and rising over time. How long are you willing to go hungry supporting capitalism in that environment, and why?
The food is right there, fully automated, the only reason you're not getting it is because you refuse to work in the mines 14 hours a day to get it. Capitalism will survive for a while but people will eventually revolt.
Not in the slightest. Government controls all the state owned companies and they even still have their five year plans and everything. So no, it isn't. Just like the Soviets and North Korea and Vietnam and ... they let the plebs fight over the last 0.1% of the economy with tokens when it suits them.
Oh right I'm so sorry. There is no true communism anywhere on the planet, nor has there ever been and therefore your precious ideal cannot be judged. Of course, if any of it were to actually be implemented, it would rapidly evolve into "real" communism (as in existing in the real world as opposed to in theory), and it would resume destroying lives like every other real communism has always done ...
But your precious ideal really is perfect (as in I don't want to argue about the ideal communism, a figment of people's imagination, because it's about as relevant as the very intricate arguing online on the politics of "magic for all ponies", the political slogan of the famous show)
But of course you want to argue the ideal since it's really about just using the power of the state to get you a free playstation and a big apartment in the center of a perfect city. Reality is that if such a thing were attempted, playstations would rapidly cease to exist, big apartments would be reserved for party "business" (and mistresses), and the city would be about as perfect as moscow was in the 1970s.
Oh, think it will/can be worse than that. The collapse of the present system can mean all out genocidal wars, either by eugenic and deranged humans doing it to themselves or true AGI and its robots removing human interference with its goals and Earth's resources.
As for the AGI and robot part, various oligarchs are talking about manufacturing millions to even billions of robots. Combine that with billions of AI agents in massive data centers all over the world. Bringing forth a new self-sustaining artificial life form that can do what humans do (if not vastly better) and at comparable numbers or mass.
Intellectual property has rarely ever been "the golden goose", despite the anecdotes of the few Henry Fords that exist.
Throughout history, the golden goose has typically been owning capital, and having market reach (see: advertising, the silk road).
No matter how amazing your IP is, if you can't produce it at economically viable scales, and if you can't put it in front of people who will buy it, then you have nothing. So venture capital comes along, buys you out, and takes all your future profits in exchange for the "risk".
On the other side of the coin, there's open source software developers, PhD's publishing papers for free, and your local musicians that like to play at events just for funsies.
I mean, imagine the kind of world we'd live in if medicine were focused strictly on profitable moats instead of open research, and... oh. Right.
Your question presupposes new algorithms must come from someone toiling. But AI will be just as capable at developing new algorithms as ripping off the existing ones, surely?
This was a huge problem before AI due to Amazon and global internet marketplaces. Why spend R&D dollars developing an amazing product when Chinese knockoffs will show up 6 months later? I wonder how many great consumer products don't exist because there's no way to recoup your R&D costs and develop a healthy market.
Imagine if there were no intellectual protections for pharmaceuticals. The government would have to take over R&D.
people keep talking about AI killing this or that, but I feel they will create new industries. 50 years ago new tech can shift industries over decade or so, but AI will shift it in 1/5 or in a year, 2 even 1/10 of the time.
I also have found that AI has not lived up to many of its promises and have dialed back what AI gets control of. My projects were turning into unmaintainable messes. The people who say coding is solved aren't paying attention.
Yes. I started two projects with AI from scratch. Both abandoned, complete mess. Projects without AI are so easy to manage, maintain, add/remove features, etc. I use AI as a search engine on my projects instead of Google. I ask what's wrong with my code and change or improve it myself based on my experience.
I have the opposite experience. I don't have the patience sometimes to clean up my code and stick to coherent conventions and organization even though the will is there. With my AI projects I watch it and if the AI starts drifting I ask it to go through and look for convention/directory structure violations and it happily cleans everything up in about 10 or so minutes.
Are you in Python by chance? Python has a lot of crazy hidden/inexplicit/spooky action at a distance stuff (especially in the frameworks) that can make LLMs gunk up code by defensively programming or just burn context chasing data provenance
I've been a vibe-coding skeptic for years, but because of the math breakthroughs of the past few weeks I decided to experiment with the latest models on some test projects. They're a lot more capable than I thought they would be. I agree that it's easy to create an unrecoverable mess, especially when you're one-shotting a lot of features without detailed instructions. But I find that as long as I'm strict about the API boundaries and force the agent to work in small chunks, it's pretty effective. As one example, I got it to write an SVG renderer in a few hours (not the whole spec, but most of the path features and text rendering), which would have taken me at least a week just for the coding part, plus extra time to learn the algorithms.
It didn't copy any source code from any external projects. I had it write a stratified sampling renderer for ground truth, then had it implement feature by feature by matching the pixels. Unless you mean it "copied" it in the sense of third-party code being part of the training data. I don't think that definition of "copy" makes any sense given how these models represent embeddings. It would also imply that humans are "copying" the things they've learned from.
I don’t think we should hold humans and models to the same standards. Humans have a very small working memory. Most humans cannot reproduce code they wrote even a year back exactly.
A model can reproduce large swaths of its training data exactly. It’s a different algorithm that powers its learning process (it’s why it needs trillions of tokens to even learn basics of language).
If there was a spectrum from copying on one end to creative production inspired from something else on the other end, the human generally lies heavily on the right end, while the model is much more on the left, that gap is large enough, that yes the model is in some sense “copying”.
Humans are also capable of memorizing long texts and sequences of numbers. It's more a question of scale than a fundamental difference. The point is that LLMs are like human brains in the sense that they're not databases that store text, even though they can and do memorize things.
Just because a model can reproduce parts of its training set doesn't mean that's what it's doing when it solves a programming problem. It can also reason about the problem, draw on its knowledge of algorithms and data structures, write tests targeting APIs it's never seen before, generate synthetic data and run experiments, etc. etc. Also, the claim that it can reproduce large swaths of its training data verbatim is an empirical one. I would be surprised if it could even reproduce 0.1% of the books it's ingested, for example.
Saying that AI can't do anything but copy or steal from humans seems to be a rhetorical technique used by people who are still unaware or in denial about the capabilities of the agentic systems released in the past few months. They can now one-shot theorems and programming problems in a few minutes that would be difficult and time-intensive for even the 99.9th percentile human expert.
Empirically I just asked Astra without using the internet to exactly reproduce the first paragraph from Hamlet, Othello, Pride and Prejudice, Sartor Resartus by Thomas Carlyle), Sybil by Benjamin Disraeli (1845). GPT-6 got everything correct:
https://chatgpt.com/share/6ac2a72c-e8bc-83e8-8f42-da8c7542de...
No human on earth can copy at this scale. Yes AI is beyond a database lookup, it does have reasoning on top of this knowledge, but my original claim that if there is a spectrum between copying with minimal changes and creative inspiration with minimal copying, AI is one the copying end while humans are on the creative end. Humans are very bad at reproducing anything verbatim, even if they wrote it like a week ago.
Obviously the scale of retrieval from memory is different, I'm not disputing that. The empirical question I was referring to is what percent of its training set the model is capable of reproducing. What looks like "large swaths" of text to us could be less than one in a million for all we know.
I'm skeptical that there is a single spectrum like you're describing. It's not well defined. Say a human and an LLM prove a new theorem independently (without external help, i.e. from their own neural weights and reasoning). How do we measure how much each of them copied from previous work, as opposed to having learned from or been influenced by it?
Sure, happy to provide you with an example of how to hold it (turns out Steve was right) =D
https://github.com/NousResearch/hermes-agent is 99% (just a guess) LLM generated. 1140 closed pull requests this week. 1.5k closed issues. The github insights page for commits doesn't load for me presumably because it can't handle this scale of commits. But I estimate ~1K commits per day on average.
You're using quantity metrics to answer a quality question.
I had a look at the kind of issues that are reported at that project (there's 15k of them, so I can at best assess a couple). It looks like a complete mess: A lot of concurrency and resource mismanagement issues and edge cases that in a better-managed project would have been avoided by construction. They will now will likely be solved by more defensive programming, driving overall complexity ever upwards.
If you really want to check some quantity metrics to try to reason about code quality, look at whether "fix" PRs are overall LOC neutral or negative (not counting tests). In this project, almost every "fix" is an addition. Worse, almost every fix is more branching.
If almost every PR is some sort of fix, and most of them add branching, and there's thousands of them weekly... That leads to only one place and I want to be nowhere near it.
I find it funny that the original complaint was: "AI made a mess of the codebase".
Your response was: "Well you're not doing it right, but these hermes devs know what they're doing".
But the blog post you linked to shows their prompt, which is:
> I want god files broken up. I want simplification across the board. I want unification of helpers and methods that can be reused. I want less if-if-if-if-if-if-else routing. I want code legibility up. I want interpretability of the codebase and how things connect to each other up.
So it sounds like AI made their code a mess too. They then tried to make the point of how much money they saved cleaning up the code with AI, that AI made a mess of to begin with.
And if you look at the merged PRs on that project, a ton of them are bug fixes... to the code the AI wrote. And that's been my personal experience too: AI creates a huge amount of churn in a codebase. Just vast amounts of PRs fixing code that the AI itself wrote.
I’ve had the best luck by spending quite a bit of time going over the big picture architecture up front and then diving into the modules to further refine the details, making sure to generate step-by-step chunks of work in Markdown format for implementation. I’ll spend literally a couple of days doing this before starting any coding.
Edit: My latest project is all GPT-6 Astra High. It takes a lot of steering to keep it from adding a bunch of, while useful, features that are not strictly enough to the point. That main issue is it’ll use a lot of extra tokens in the process!
Do you mind sharing any code from what you have produced? People talk about LLM successes and failures, but what's there to really talk about when the code can speak for itself?
In case it is unclear, I am genuinely curious. I have great success with chatbots, but vibing coding has never gotten me further than a proof-of-concept.
An expert doesn't waste their time going on an ideological crusade against tooling. Even the most ardent anti-AI people are admitting it is useful for programming. Eventually you will have to do the same. The more you dig your feet in the worse it's going to be for you when that day comes.
I've adopted open source projects and have been working exclusively in languages I have very little experience with: typescript and go.
I'm using local models, and they go slow enough that I have no trouble following along with what they're doing; but visually, both go and typescript, along with react native, make me puke. So I wouldn't be able to do this without AI.
I describe how to do it in my comment history, but it's basically a Super-TDD along with some custom engineering harness.
I don't want to say skill issue, but the same way you can give a chain saw to a teenager and one to a skill craftsman, well, AI can obviously create whatever you want it to do.
I think some of the variety is simply how fast SOTA models pump out garbage that you simply have to close your eyes because it's not sensible to just watch characters flow across the screen.
Almost all the coding I'm doing via AI is just faster than readable. But I can see the thinking traces and I stop to model when it's obvious it doesn't understand my intent, etc.
So I'm not doubting you created garbage. I'm just doubting that it's a product of soley AI use.
I've had a different experience. My code is better. The ease of refactoring and practicing very defensive programming without getting burned out from the repetitive boilerplate. It all comes together better than it did when I was programming professionally too many moons ago. IFF I don't let the LLM toddler run amok. :-)
Seriously, I find I need to slow down the rate of change. I don't move forward until I understand the change proposed and have updated the docs. At the same time, I find that keeping up with the LLM/agent is exhausting. 3 hours with an LLM leaves me as tired as 6 hours with a keyboard had previously. I find that coding when tired or fuzzy yields code that shouldn't have been written in the first place. Sadly, once it's been written and debugged, the temporary fix becomes permanent.
I'm reading y'all comments and it seems we're still finding our footing, and will for some time. I have the luck that I have access to pretty much all frontier and other models alike. I've been extremely negatively biased towards any LLM use in the start. Then the influx from juniors came, then the revolt of doing PRs on such slop came, then some structured methods how to do LLM work came, then vibe projects came, etc, etc. I literally have all the described experiences you've guys mentioned. From good, to bad, to ugly. It's like there's no one particular way about doing this, and no two projects share the same approach - just like ye olde times.
People, especially those who don't do software development, often conflate coding and software development.
The models aren't good at architecture and design. But they take direction on architecture and design and design well and can refactor code quite effectively. AI agents can absolutely be used to clean up vibe coded code bases once you figure out if the investment is worth it. The mess can be avoided if you give them sufficient guidance on architecture and design upfront.
That said, doing so purely in text form doesn't feel great right now. I've been thinking about UML lately. The problem with that was the roundtrip after the code was generated and then the implenetation happened. I don't necessarily think UML is the solution, but neither is walls of dense text.
>The models aren't good at architecture and design.
Can you elaborate to back up this claim? WHat exactly is your yardstick for "being good at SW design and architecture"?
Because I found the current SOTA AI models being great at architecture and design, much better in fact than most average real-world devs. Is your yardstick just the John Carmacks of the world by any chance? Because most devs are not John Carmack. They are also not Linus Torvalds, they are not Stallmann, etc.
Maybe your LLM experience is still stuck in the 2023 era of ChatGPT?
I'm currently driving my AI coding agent harness through the process of refactoring itself. This is the last big refactor I need done before I can polish and release it as OSS later this month. So its not a large scale project I'm working on at present. The general problem is that mostly add new code and try to minimize editing existing code, which effectively means the code base is grown organically rather than being intentionally architected and designed. A few of specifics:
1) duplication - LLMs are great at generating lots of text, so its faster and easier for them to generate entirely new facilities that overlap heavily with existing ones then it is for them find existing facilities that should be expanded and refactored (note I just said 'find'; actually editing raises the time and difficulty even more). This is fine for a while as the duplicate facilities usually work just fine, up until something needs to be changed across all of them and they miss changing one or more of them, things break, and a bunch of tokens have to be burned tracking down the issue.
2) ever increasing surface area - even when making changes that do expand a facility without much duplication they frequently only add without removing much of anything or changing the overall design of the facility to reduce the amount of state its tracking and the number of branches it has based on that state. I've never seen one decide to split up something large or with too many responsibilities on their own. They will happily create a god class or function and just keep making it bigger.
Oh my god. I am sad to have to acknowledge this even if the models have “theoretically” gotten better. I am not a programmer but I am super opinionated with the design and structure of the code and need to make sure whatever I write/ask to generate and use for my own tasks is understood by me at least once. I wasn’t this way in 2023.
I am constantly reading/trying for an agent to help me write simple code without spending too much time. But so far nothing has worked. I wish someone figures it out otherwise, I personally would not be able to realize the AI agent productivity benefits that others are supposedly seeing.
If you specify a structure, pattern, or design, it will stick to that design after a code review phase. Just tell it what standards you have, and after two rounds, it will have written what you described.
I've had the opposite experience. Spec driven development really makes things much easier to maintain. If you're just yolo'ing it and throwing prompts around things will fall apart fast.
The people saying programming is solved couldn't program in the first place is what I've noticed. So to them it really does feel solved, things "work" and they don't have to learn what they don't know about programming.
I've been programming professionally for decades. LLMs are extremely useful. At this point if you haven't figured out how to get value out of them, you're either holding them very wrong or you're being willfully ignorant.
I think it depends which LLM tool you're using. If you're using an older, worse model (the kind that you can use for free), the experience is significantly more frustrating. On the other hand, I'd say that the current best models are very useful with a skilled operator.
There are a couple of notable counterexamples here (nobody sane thinks Carmack doesn't know how to program, for example), but by and large I agree with your observation. The people excited about programming with LLMs are, on average, people who weren't good at programming to begin with. Still, given that these counterexamples do exist I try to avoid painting with an overly broad brush for the sake of nuance.
As usual, when people post comments like this they never specify what model and when they used it. There is a huge difference between GPT 3 and 6 for example.
Late 2025 also had a step change when agents could largely code autonomously without handholding like previously, and to be honest it's not worth hearing opinions about AI from before that time, that's how significant the change was.
Were you reading through the code it was generating to make sure the flow was intuitive and comprehensible for each PR? Projects only turn in to maintainable messes if you blindly merge in unmaintainable messy code.
"Coding is solved" is a very high bar, and current-gen AI/LLM is very far from that.
Where AI did make alot of impact is triage. I can throw a messy bug report of an intermittent issue at claude, and tell I need issue reproduced and fix developed, and there is a well above 50% chance it'll deliver. Never commit that fix to the codebase as-is, of course.
I treat AI like an intern or junior team member. With enough guidance, they can contribute a lot but you can't let them loose without supervision or they usually will produce a big mess. As of now, you are still responsible for overall architecture. One strong indicator that something is going wrong are big pull requests where the AI has rewritten large sections of the code.
AI is a useful tool but it absolutely needs a lot of human guidance.
Unlike a compiler it won't give up at the first sign of trouble but that just means it left alone it will dig bigger and bigger holes.
Treat prompt engineering as a discipline and refine your technique. When it produces garbage throw out the work and start over until you figure it out.
Were you reviewing the changes? I've seen lots of projects with this problem. But if AI PRs have to pass the same review bar as any other PR then it shouldn't be an issue in theory (if you can actually maintain the discipline).
Isn't it tiring to keep up with considering the speed of the generated code? And if you want to be careful with the review you lose a significant part of the speed advantage.
It's the same as any other PR. Keep the changes small and contained to that feature. AI can do this if you instruct it to, there is no need to vibe code some 40k line monstrosity.
If dev approaches zero, but you review at the same pace as you always have, are you in a better position? Yes.
Will you potentially have a backlog of code waiting for review? Also yes.
Would you prefer to be waiting for the dev team for all of the time instead, then still have the same amount of reviewing to do at the end of it? Absolutely not.
Yes, but it's the only option if you want to retain a maintainable code base. And you can always slow down. You'll probably still be a bit faster than you were before.
Vibe coding has always had and always will have one fundamental flaw: you didn't write the code it produced, therefore you don't have a good understanding/mental model of it.
On the other hand asking these clankers "review the feature branch I wrote" and "review my entire codebase for bugs" or "help me debug this" has saved me months of prospective work.
And more recently most major models have been getting _really_ good at RE, for example you can have OAI models (and maybe A/'s if they don't refuse) use idalib MCP and reverse-engineer stuff from start to finish, then follow up with GLM 5.3 for vuln assessment and exploit PoC.
Stuff that used to take weeks or months now just takes a few hours, or less.
I am not saying you are right or wrong, but I don’t think you can make such a broad conclusion simply because of your own experience.
You say you used AI and your projects turned into unmaintainable messes, so your conclusion is that it means AI is not living up to its promises.
I guess if the argument is “AI makes it so you always get a great result no matter how you use it”, then your argument is sound. Your projects not working out proves that AI doesn’t always work no matter what.
However, that doesn’t mean you can’t use AI to create sustainable and well organized code. Failing to do something doesn’t mean it’s impossible and anyone who thinks they can is not paying attention.
I can’t run a marathon. If I went out and tried to run one, I would get a few miles and collapse, failing completely.
I don’t think it would be reasonable, though, at that point to say “running a marathon is impossible, anyone who says they can do it clearly lying. I tried and didn’t even make it 5 miles!”
I wish people would stop assuming their experience with something is the only possible truth.
Stock market dynamics are fundamentally different now compared to the last period of sustained high inflation (70s)... And stocks have done fantastically over the last few years of stubborn inflation.
The roaring 20s were driven by a money cycle formed with post WWI farm machinery foreign demand, US factory growth to meet that stimulated demand, stocks to fund factory export growth market, and debt generated by unsecured customer financing.
Eventually the cycle/bubble collapsed, and the same 1930s tariff policy literally starved people for almost a decade.
I really hope we don't repeat that history. However, the Bear market distant roar can't be ignored. =3
If they, through standard accounting practices, cannot show they are profitable how do we know they are profitable? I assume they can't just say "we are profitable assuming you ignore the costs we paid to build up XYZ". I'm guessing they could pre - rent compute for 10 years or something and maybe that's all on the balance sheet but you'd just amortize that and I think that falls under GAAP?
Intellectual property incentivized many and laid many golden eggs, but why would someone toil to create a new algorithm to solve a problem if someone else is just going to steal it in hours?
But who knows, maybe imploding the software industry is a good thing. I may lose my house in the process, but I think we could all get back to touching a little more grass.
reply