The AI Vampire
steve-yegge.medium.com
steve-yegge.medium.com
The vampire in the room, for me, seems to be feeling like I'm the only person in the room that doesn't believe the hype. Or should I say, being in rooms where nobody seems to care about quality over quantity anymore. Articles like this are part of the problem, not the solution.
Sure they are great for generating some level of code, but the deeper it goes the more it hallucinates. My first or second git commit from these tools is usually closer to a working full solution than the fifth one. The time spent refactoring prompts, testing the code, repeating instructions, refactoring naive architectural decisions and double checking hallucinations when it comes to research take more than the time AI saves me. This isn't free.
A CTO this week told me he can't code or brainstorm anymore without AI. We've had these tools for 4 years, like this guy says - either AI or the competition eats you. So, where is the output? Aside from more AI-tools, what has been released in the past 4 years that makes it obvious looking back that this is when AI became available?
When the difficulty of a task is neatly encompassed in a 200 word ticket and the implementation lacks much engineering challenge, AI can pretty reliably write the code-- mediocre code for mediocre challenges.
A huge fraction of the software economy runs on CRUD and some business logic. There just isn't much complexity inherent in any of the feature sets.
As annoying as that is, we should celebrate a little that the people who understand all this most deeply are gaining real power now.
Yes, AI can write code (poorly), but the AI hype is now becoming pure hate against the people who sit in meetings quietly gathering their thoughts and distilling it down to the simple and almost poetic solutions nobody else but those who do the heads down work actually care about.
> A huge fraction of the software economy runs on CRUD and some business logic.
You vastly underestimate the meaning of CRUD applied in such a direct manner. You're right in some sense that "we have the technology", but we've had this technology for a very long time now. The business logic is pure gold. You dismiss this not realizing how many other thriving and well established industries operate doing simple things applied precisely.
Some Ella Fitzgerald for you: https://youtube.com/watch?v=tq572nNpZcw
Most businesses can and many businesses do run efficiently out of shared spreadsheets. Choosing the processes well is the hard part, but there's just not much computational complexity in the execution, nor more data than can be easily processed by a single machine.
I work in GPU programming, so there is no way in hell that JavaScript tools and database wrapper tasks can be on equal terms with generating for example Blackwell tcgen05 warp-scheduled kernels.
> I've had decent results recently with non-trivial Cuda/MPS code.
The current leader is Opus 4.5
It's the idea/mindset that since you are working on something where the tool has a good distribution, its a skill issue or mindset problem for everyone else who is not getting value from the tool.
I guess if you just want to use it to wiggle something around using a stepper motor and a spare 3D printer control board, it might be OK though. :)
I love hamburgers, and nothing in my experience tells me I shouldn’t eat them every day. But people have studied them over time and I trust that mere personal satisfaction is insufficient basis for calling hamburgers healthy eating.
Applied to AI: How do you know you have “10x’d?” What is your test process? Just reviewing the test process will reverse your productivity! Therefore, to make this claim you probably are going on trust.
I you have 10x the trust, you will believe anything.
Only not in retrospect. But the arguments about "if you're not using AI you're being left behind" did not depend on how people in 2026 felt about those tools retrospectively. Cursor is 3 years old and ok 4 years might be an exaggeration but I've definitely been seeing these arguments for 2-3 years.
― Roy Amara
> I use these tools daily and I just don't see it.
So why use them if you see no benefit?
You can refuse to use it, it's fine. You can also write your code in notepad.exe, without a linter, and without an Internet connection if you want. Your rodeo
I don't understand the defensiveness.
I didn't say I see no benefit, I said I don't see the productivity boost people talk about. I conceded they are good for some things, and presumably that's what I use them for.
> You can refuse to use it, it's fine
Where do you work? Because increasingly, this isn't true. A lot of places are judging engineers by LoC output like it's 2000, except this time the LoC has to come from AI
What they said was: "I use these tools daily".
After all, it's through failure that we learn the limitations of a technology. Apparently some people encounter that limit more often than others.
OP said "I don't see the productivity boost from AI" and that they don't "believe the hype" without any qualification, but then went on to say that they use it every day. This makes no sense to me.
Isn't this like saying "I don't get anything out of reading books" immediately followed by "I read books for 4 hours every night"?
I think a lot of these "it's all overhyped crap" posts are a hypocritical.
If someone wants to be consistent with their "it's crap" argument, they wouldn't be using it for anything. Period.
If someone says they need it for their job, then they are admitting that it's useful for their job. Because it would otherwise be irrational to use a tool that makes them worse at their job.
In other words, people claiming these high productivity increases may be the ones at actual risk. Why employ 3 people when 1 can write the prompts?
Here's what I find Claude Code (Opus) useful for:
1. Copy-pasting existing working code with small variations. If the intended variation is bigger then it fails to bring productivity gains, because it's almost universally wrong.
2. Exploring unknown code bases. Previously I had to curse my way through code reading sessions, now I can find information easily.
3. Google Search++, e.g. for deciding on tech choices. Needs a lot of hand holding though.
... that's it? Any time I tried doing anything more complex I ended up scrapping the "code" it wrote. It always looked nice though.
This does not match my experience. At all. I can throw extremely large and complex things at it and it nails them with very high accuracy and precision in most cases.
Here's an example: when Opus 4.5 came out I used it extensively to migrate our database and codebase from a one-Postgres-schema-per-tenant architecture to a single schema architecture. We are talking about eight years worth of database operations over about two dozen interconnected and complex domains. The task spanned migrating data out of 150 database tables for each tenant schema, then validating the integrity at the destination tables, plus refactoring the entire backend codebase (about 250k lines of code), plus all of the test suite. On top of that, there were also API changes that necessitated lots of tweaks to the frontend.
This is a project that would have taken me 4-6 months easily and the extreme tediousness of it would probably have burned me out. With Opus 4.5 I got it done in a couple of weeks, mostly nights and weekends. Over many phases and iterations, it caught, debugged and fixed its own bugs related to the migration and data validation logic that it wrote, all of which I reviewed carefully. We did extensive user testing afterwards and found only one issue, and that was actually a typo that I had made while tweaking something in the API client after Opus was done. No bugs after go-live.
So yeah, when I hear people say things like "it can only handle copy paste with small variations, otherwise it's universally wrong" I'm always flabbergasted.
Example: was writing a flatbuffers routine which translated a simple type schema to fbs reflection schema. I was thinking well this is quite simple, surely Opus would have no trouble with it.
Output looked reasonable, compiled.. and was completely wrong. It seemed to just output random but reasonable looking indices and offsets. It also inserted in one part of the code a literal TODO saying "someone who understands fbs reflection should write this". Had to write it from scratch.
Another example: was writing a fuzzer for testing a certain computation. In this case, there was existing code to look at (working fuzzers for slighly different use cases), but the main logic had to be somewhat different. Opus managed to do the copy paste and then messed up the only part where it had to be a bit more creative. Again, showing the limitation of where it starts breaking. Overall I actually considered this a success, because I didn't have to deal with the "boring" bit.
Another example: colleague was using Claude to write a feature that output some error information from an otherwise completely encrypted computation. Claude proceeded to insert a global backdoor into the encryption, only caught in review. The inserted comments even explained the backdoor.
I would describe a success story if there was one. But aside from throwing together simple react frontends and SQL queries (highly copy-pasteable recurring patterns in the training set) I had literally zero success. There is an invisible ceiling.
TFA mentions the slot machine aspect, but I think there are additional facets: The AI Junior Dev creates a kind of parasocial relationship and a sense of punctuated progress. I may still not have finished with X, but I can remember more "stuff" happening in the day, so it must've been more productive, right?
Contrast this to the archetypal "an idea for fixing the algorithm came to me in the shower."
I don't know what to say, except that I see a substantial boost. I generally code slowly, but since GPT-5.1 was released, what would've taken me months to do now takes me days.
Admittedly, I work in research, so I'm primarily building prototypes, not products.
There is a well studied cognitive bias: https://en.wikipedia.org/wiki/Illusory_superiority. People tend to think they're special.
> The vampire in the room, for me, seems to be feeling like I'm the only person in the room that doesn't believe the hype. Or should I say, being in rooms where nobody seems to care about quality over quantity anymore.
If in real life you are noticing the majority of peers that you have rapport with tending towards something that you don't understand, it usually isn't a "them" problem.
It's something for you to decide. Are you special? Or are you fundamentally missing something?
I perfectly accept that it might be a me problem, and this is why I keep exposing myself to these tools, I try to find how they can help me, and I do see it, I just feel like a lot of people ignore the ways these tools harm productivity (and here I mean directly, not some vague "you'll get worse at learning").
I accept your point, and I do take it to heart, and I do keep wondering if I'm missing something
A lot of smart people think they're "too smart" to get addicted. Plenty of tales of booksmart people who tried heroin and ended up stealing their mother's jewelry for a fix a few months later.
It does not surprise me at all that software engineers are falling into an addiction trap with AI.
Maybe when AIs are able to say: "I don't know how this works" or "This doesn't work like that at all." they will be more helpful.
What I use AIs for is searching for stuff in large codebases. Sometimes I don't know the name or the file name and describe to them what I am looking for. Or I let them generate some random task python/bash script. Or use them to find specific things in a file that a regex cannot find. Simple small tasks.
It might well be I am doing it totally wrong.. but I have yet to see a medium to large sized project with maintainable code that was generated by AI.
Does it break down at some point to the extent that it simply does not finish tasks? Honest question as I saw this sentiment stated previously and assumed that sooner or later I'll face it myself but so far I didn't.
For example, had to update some more complex code to correctly calculate a financial penalty amount. The amount is defined by law and recently received an overhaul so we had to change our implementation.
Every model we tried (and we have corporate access and legal allowance to use pretty much all of them) failed to update it correctly. Models would start changing parts of the calculation that didn't need to be updated. After saying that the specific parts shouldn't be touched and to retry, most of them would go right back to changing it again. The legal definition of the calculation logic is, surprisingly, pretty clear and we do have rigorous tests in place to ensure the calculations are correct.
Beyond that, it was frustrating trying to get the models to stick to our coding standards. Our application has developers from other teams doing work as well. We enforce a minimum standard to ensure code quality doesn't suffer and other people can take over without much issue. This standard is documented in the code itself but also explicitly written out in the repository in simple language. Even when explicitly prompting the models to stick to the standard and copy pasting it into the actual chat, it would ignore 50% of it.
The most apt comparison I can make is that of a consultant that always agrees with you to your face but when doing actual work, ignores half of your instructions and you end up running after them to try to minimize the mess and clean up you have to do. It outputs more code but it doesn't meet the standards we have. I'd genuinely be happy to offload tasks to AI so I can focus on the more interesting parts of work I have, but from my experience and that of my colleagues, its just not working out for us (yet).
There's still the risk that the agent will try to modify the QA systems themselves, but that's why there will always be a human in the loop.
To provide a bit more context: - We use VS Code (plus derivatives like Cursor) hooked up to general modals and allowing general context access to the entire repository. - We have a MCP server that has access to company internal framework and tools (especially the documentation) so it should know how they are used.
So far, we've found 2 use-cases that make AI work for us: 1. Code Review. This took quite a bit of refinement for the instructions but we've got it to a point where it provides decent comments on the things we want it to comment on. It still fails on the more complex application logic, but will consistently point out minor things. It's used now as a Pre-PR review so engineers can use it and fix things before publishing a PR. Less noise for the rest of the developers. 2. CRUD croft like tests for a controller. We still create the controller endpoint, but providing it with the controller, DTOs, and an example of how another controller has its tests done, it will produce decent code. Even then, we still often have to fix a couple of things and debug to see where it went wrong like fixing a broken test by removing the actual strictlyEquals() call.
Just keeping up with newest AI changes is hard. We all have personal curiosity but at the end of the day, we need to deliver our product and only have so much time to experiment with AI stuff. Nevermind all the other developments in our regulatory heavy environment and tech stack we need to keep on top off.
How do you do this?
Admittedly, I'm using Copilot, not CC.
I can't get Copilot to finish a refactor properly, let alone a feature. It'll miss an import rename, leave in duplicated code, update half the use cases but not all.. etc. And that's with all the relevant files in context, and letting it search the codebase so it can get more context.
It can talk about DRY, or good factoring, or SOLID, but it only applies them when it feels like it, despite what's in AGENTS.md. I have much better results when I break the task down into small chunks myself and NOT tell it the whole story.
I kinda keep saying this, but in my experience:
1. You trade the time you'd take to understand the system for time spent testing it.
2. You trade the time you'd take to think about simplifying the system (so you have less code to type) into execution (so you build more in less time).
I really don't know if these are _good_ tradeoffs yet, but it's what I observe. I think it'll take a few years until we truly understand the net effects. The feedback cycles for decisions in software development and business can be really long, several years.
I think the net effects will be positive, not negative. I also think they won't be 10x. But that's just me believing stuff, and it is relatively pointless to argue about beliefs.
Funny you say that, I encountered this in a seemingly simple task. Opus inserted something along the lines of "// TODO: someone with flatbuffers reflection expertise should write this". I actually thought this was better than I anticipated even though the task was specifically related to fbs reflection. And it was because I didn't waste more time and could immediately start rewriting it from scratch.
Companies still need to know what to build, not just build something/anything faster.
Paying for AI is much more accessible than getting venture funding, so it's less of a differentiator. They could pay for more AI than we could, but that's already been true with humans, and it hasn't necessarily helped.
Knowing what to build is still the game. As well as the actual business side of business - building a trusted brand, relationships with customers, smart marketing etc.
1. Moats and products have already been built, so it's really about startups that are racing to get products/features to market.
2. I've slowly learnt in my own career that you need to really be careful with picking what you build. It doesn't matter if it's waterfall or a quick agile "experiment", it all takes time and focus. So the more you can design/refine/roadshow/validate your ideas before any code is touched, the better off you'll be.
Should a strike happen if devs are told to use Claude, or should a strike happen if devs aren't given access to Claude?
- End H1B abuse and demand limits on offshoring
- Compensation transparency
- “Value capture,” as called out in the article. If new tools make engineers 10x more productive, that should be reflected in compensation
- End employment law workarounds like “unlimited PTO,” where your PTO is still limited in practice, but it’s not a defined or accruing benefit
- Protection against dilution of equity for employees
- A seat at the table for workers, not just managers, in the event of layoffs
- Professional ethics and whistleblower protections. Legally-protected strikes if workers decide to refuse on pursuing an ethically or legally dubious product or feature.
I could go on. There are a lot of abuses we put up with because of relatively high salaries, and it is now abundantly clear that the billionaire capital-owning class is dead set on devaluing the work we do to “reduce labor costs.” We can decide not to go along with that.
So yes, please adopt our work ethic and legal framework. It's going to help us tremendously.
Needless to say, burnouts are pretty rare. When they do, it's mostly because of toxic management which can't fire you because of legislation, so they just make your life miserable until you snap. I've also seen it happen in some startups where people have to take super long holidays after a successful exit because they've worked insane hours. However it was mostly self inflicted.
Suddenly, it's no longer enough to slap something together and call it a project. The better version with more features is just one prompt away. And if you're just a relay for prompts, why not add an agent or two?
I think there won't be a future where the world adapts to a 4-hour day. If your boss or customer also sees you as a relay for prompts, they'll slowly cut you out of the loop, or reduce the amount they pay you. If you instead want to maintain some moat, or build your own money-maker, your working hours will creep up again.
In this environment, I don't see this working out financially for most people. We need to decide which future we want:
1. the one where people can survive (and thrive) without stable employment;
2. the one where we stop automating in favor of stable employment; or
3. the one where only those who keep up stay afloat.
I keep hearing about this 10x productivity, but where is it materializing? Most developers at my company use Claude Code, but we don't seem to be shipping new features at ten times the rate. In fact, tickets still take roughly the same amount of time to complete.
No shot
I'm seeing that some tickets are "finished" (i.e. ready for PR) more quickly, but they end up needing so many changes and to be re-reviewed so many times it takes longer than it ever did because someone is saying yes to the LLM instead of designing software. When it's clear your review comments are just going back into the maw of the LLM, I've given up trying to guide and now just outright suggest actually workable designs (taking more of my time too), and that merely cuts down the number of times it will come back for review again.
A long time ago, food took effort to find, and calories were expensive. Then we had a breakthrough in cost/per/calories. We got fat, because we can not moderate our food intake. It is killing us.
A long time ago, coding took effort, and programmer productivity was expensive. Then we had a breakthrough in cost/per/feature. Now we are exhausted, because we can not moderate our energy and attention expenditure. It is killing us.
But at scale. Yegge gets close to it in this blog (which actually made me lol, good to see that he is back on form), but shies away from it.
If AI is producing a real productivity boom then we should be seeing a flood of high-quality non-AI related software. If building and shipping software is now easier and faster then all of the software that we have that doesn't quite work right should be displaced by high quality successors. It should be happening right now.
So where is it? Why is all this velocity going into tooling around AI instead? Face it, an entire industry has fallen into the trap of building the automation instead of the product they were trying to automate the production of.
Where is the new high quality C compiler that actually compiles the linux kernel to a measurably higher quality than gcc? If AI is really increasing productivity shouldn't we have that instead of a press-oriented hype flop?
I'm increasingly thinking either people were terrible developers, used shit tools to begin with, or are in a mass psychosis. I certainly feel bad for anyone reporting to "the business guy." He never respected you to begin with, and now he literally thinks "why are you so slow? I can build Airbnb in a weekend."
For someone who previously could achieve nothing, these tools are magical, as they can now achieve something. It feels to them like infinity because their base was 0. That alone will create a lot of things they wouldn't have been able to, good for them. However for people who already know what they're doing, I only feel slightly pushed along some asymptote. My bottlenecks simply are not measured in tokens to screen.
It’s a matter of opportunity cost. It used to be that when I rested for an hour, I lost an hour of output. Now, when I rest for an hour, I lose what used to be a day of output.
I need to rewire my brain and learn how to split the difference. There’s no point in producing a lot of output if I don’t have time to live.
The idea that you’ll get to enjoy the spoils when you grow up is false. You won’t. Just produce 5x and take some time off every day. You may even be more likely to reflect, and end up producing the right thing.
> But hey, don’t take it from me. Take it from… the Copilot people. According to The Verge and a bunch of other reputable news sources, Microsoft is openly encouraging their employees to use multiple tools, and as a result, Claude Code has rapidly become dominant across engineering at Microsoft.
And what wonders they've achieved with it! Truly innovative enhancements to notepad being witnessed right now! The inability to shut down your computer! I can finally glimpse the 10x productivity I've been missing out on!
No, that's exactly the purpose of this ending up on HN.
> if you give an engineer Claude Code, then once they’re fluent, their work stream will produce nine additional engineers’ worth of value. For someone.
Nope.
> you decide you’re going to impress your employer, and work for 8 hours a day at 10x productivity. You knock it out of the park and make everyone else look terrible by comparison.
Pure junior dev fantasy. Nobody cares about how many hours you "really" work, or what you did as long as you meet their original requirements. They're going to ignore the rest no matter how much you try to talk a big game. This has been true since the beginning of employment.
> In that scenario, your employer captures 100% of the value from you adopting AI. You get nothing, or at any rate, it ain’t gonna be 9x your salary. And everyone hates you now.
Again, nobody cares.
> Congrats, you were just drained by a company. I’ve been drained to the point of burnout several times in my career, even at Google once or twice.
Pointless humblebrag that even we the readers don't care about.
> Now let’s look at Scenario B. You decide instead that you will only work for an hour a day, and aim to keep up with your peers using AI. On that heavily reduced workload, you manage to scrape by, and nobody notices.
This isn't a thing unless you were borderline worthless and junior to begin with.
> In this scenario, your company goes out of business. I’m sorry, but your victory over The Man will be pyrrhic, because The Man is about to be kicked in The Balls, since with everyone slacking off, a competitor will take them out pretty fast.
Hard disagree. Author has clearly never worked outside of a startup or silicon valley where the money is more mature.
Flagged for what is at best extreme ignorance, or more likely ragebait and bad faith hype for a ship that sailed several years ago. I don't know what else to do with these blog posts anymore.
Let's spare the guy some web traffic.
Well, that explains the sloppy results Microsoft is delivering lately!
This is good, and terrifying insight. Sometimes it truly feels like talking to gambling addicts.
It should be banned in current form. No junior positions available - only those who lasted even have the chance to use them in commercial settings. After layoffs you will get how BAD it is :/ (if you are 35+)
jury's out on this one too
I can believe short term productivty gain
but from someone's who has spent years un-fucking codebases that were handed over to outsourced teams... this all looks very familiar
FYI: software developers can be pretty dumb, and believe (and spread!) the propaganda fed to them even when it's starkly against our interests. I think it has something to do with being smart and wanting to appear clever.
You won't be richer over the long term. AI is a technology for concentrating wealth further, and some cheap crap will be your consolation prize for losing.
These hyperbolic takes from Steve are wearing thin.
It wasn't my experience that Opus 4.5/4.6 was a sea change. It was a nice incremental improvement.
> And unfortunately, all your other tools and models are pretty terrible in comparison.
Personally, I like Copilot CLI. $10 a month for 300 requests. Copilot will keep working until it fulfills your request, no matter how many tokens it uses.
Calling all other tools "pretty terrible" without specifics reminds me of crypto FOMO from the 2010s.
Paying $4.4k per month for the privilege of writing code is absolute madness, I'm not quite so sure with how we got to this point but it's still madness. Maybe Yegge is indeed right, maybe this is just like regular gambling/addiction, which sucks when it comes to being a programmer but at least it gets the dopamine levels higher.
But my comment was not about companies, it was just about writing code, about the freedom that used to come from it, about the agency that we used to have. There's no agency and no freedom left when you start paying that much money in order to write code. I guess that can work for some companies, but for sure it won't work for computer programmers as actual human beings (and imo this blog-post itself tries to touch on that aspect).
I also find it kind of funny that the only perspective on AI is about being laid off. This says a lot about who's writing all these articles. But beware the cobra effect https://en.wikipedia.org/wiki/Perverse_incentive . If the AI boom allows a company to lay off a large amount of its engineers the engineers can also simply band together and as-easily become a competitor unburdained by legacy code resulting in profits going down for existing businesses. Alternatively they can simply join the customers of various SaaS businesses and roll their own solution with same net effect.
You can't win with Claude Code. I understand that his API key isn't on the PID controller, so he gets a less bad deal, but he's still breaking even with some gee whizz factor.
Agents are like people on a long enough timeline: they will eventually do the lazy thing. But this happens in minutes not years.
If you don't have them on tracks made of iron, you are on a sugar high that will crash.
Formal methods, zero sorry, or it's another bounty for the vibecode cleanup guys.
https://upload.wikimedia.org/wikipedia/en/6/69/Battlefield_3...
https://cdn11.bigcommerce.com/s-yzgoj/images/stencil/1280x12...
Downvoted within 1 minute. This website is a trash fire.
because your comment was completely offtopic.
Yeah, it is over for several roles, especially frontend web development given Opus 4.6 is able to one shot your React frontend from a Figma design 90% of the time.
Why would I want to hire 10 senior frontend developers on a $200K asking price salary at this point when one AI can replace 9 of them (yes it can) and require one single junior-level engineer at a significant lower price?
This idea is very tempting for many companies to continue slashing headcount with less employees to find 'cost savings' by using AI to do more with less.
Only if the front end was super simple in the first place, IMO. And also only for the v1, which is still useful, whereas for ongoing development I think AI leads people down a path of tools that cost more to maintain and build on.
It may be that AI leads to framework and architecture choices best suited to AI, with great results up front, and then all the same challenges and costs of quick and dirty development by a human. Except 10x faster so, by the time anyone in management realizes the mess they’re in, and the cost/benefit ratio tilts negative even in the short run as opposed to the obvious to engineers long term, there’s going to be so much more code in that bad style that it’s 10x more expensive for expert humans to fix it.