> Software teams are capturing a lot of the speedup in the form of more leisure or quasi-leisure time. (To greatly simplify: software teams are offering 2x speedups to their companies. They're then using tools that allow them to work 10x as efficiently, and taking nice long coffee breaks.)
If the leadership (C-suite) is afraid to take the full responsibility over AI output, then the only rational choice for developers is to "pace the frontier" themselves.
> We are in fact seeing software improvement, in the form of fewer bugs and better performance.
I would like to see some of this software improvement in the form of fewer bugs. If anything it's adding more bugs.
I'll give an example from a game I like called ARC Raiders. For the first ~9 months they did not fix any of the bugs in the main menu. The damn main menu! The first thing you see when you load the game! They also failed to fix, in that time, the bug where simply pressing the Xbox button on the controller would break in-game voice chat.
They only just solved that, as of a few days ago. The main menu bugs still exist.
Dealing with third-party overlays (I presume that's the main menu bug you're talking about) is one of the most annoying things in game development due to how it hooks into the rendering pipeline and not an easy fix.
AI coding is great but the distribution channels have been closed for a while. App stores are effectively closed for new apps and old internet is dead.
> We are in fact seeing software improvement, in the form of fewer bugs and better performance.
Anecdotally, completely disagree. Maybe this is nostalgia, but I perceive a distinct and massive drop in software quality over the past few years.
Worst offenders for me are the poster children for LLM-based development. I run into bugs and performance issues with both Claude and Codex literally every day. I don't think I'm an outlier here, either.
Dev being reckless with LLMs are going to ship faster than the one using it responsibly to improve quality, so it would make sense to first see a drop in quality before things improve.
AI firm code are likely a different category though.
- more bugs becoming visible that were hidden within the software we use daily
- more bugs (in absolute terms) as more people create more software in a very short amount of time (regardless of their professional training)
- less bugs in existing software as the developers and dev orgs of existing projects is capable of automating bug hunting, test case creation, ... in essence: They get a multiplying effect out of AI regarding their capabilities.
In short: Actual developers with actual experience (the more, the merrier) can do better work while the barrier to enter "the market" for untrained individuals is a lot lower resulting in a huge amount of new software whose "success" is to be determined.
> - less bugs in existing software as the developers and dev orgs of existing projects is capable of automating bug hunting, test case creation, ... in essence: They get a multiplying effect out of AI regarding their capabilities.
I'm not convinced there are more than a handful of orgs in the world who are disciplined enough to achieve this. Or organisationally willing to increase their QA Eng headcount sufficiently to achieve it.
To add some related evidence to that, I browsed this report[1] for other reasons the other day, but it stood out to me that _100%_ of Forbes Global 2000 tech leaders said "their organization has experienced some form of AI-related downtime".
There are so many stupid little bugs in software these days that should be straightforward to fix, but just aren't fixed. Visual quirks and rendering issues. Notification bells that don't clear properly. Formatting problems. And vibecoded apps like Claude Code themselves are the biggest perpetrators.
> Software teams are capturing a lot of the speedup in the form of more leisure or quasi-leisure time. (To greatly simplify: software teams are offering 2x speedups to their companies. They're then using tools that allow them to work 10x as efficiently, and taking nice long coffee breaks.)
Writing code is not the bottleneck anymore, making decisions is. Making the right decisions needs context and thinking. A coffee break might actually help productivity. Driving a Ferrari full speed in a crowded city - only because you can - doesn't make the outcome better.
I agree with you, but I'd say that writing code hasn't been the bottleneck in a lot of industries for quite some time.
Even in the 90s, the internet and the web dramatically lowered the costs of distribution, and YC itself was built on top of that (ViaWeb).
Then there has been so many frameworks, libraries, tools, services and APIs to pick and choose from that most of the coding work has been about organization, selling an idea and doing a lot of testing. Which are still unsolved problems.
The only new idea here is the idea that Software Developers are disposable.
This was brought by the same people who did RTO and layoffs post-covid, but now with support of non-tech workers who want a piece of that (supposedly) sweet money.
And yet the clueless fans swoon about how much faster they can spit out unmaintainable code now. There are a lot of people who will have to learn the hard way.
I can't believe OP here. Where are these software teams, are they in the room with him? My teams are more stressed than ever: leadership doesn't understand why they aren't 5x more productive - even if they were 1.1x more productive, the ops teams aren't scaling, the defect management isn't scaling, the guy yelling about their Jira ticket status updates is just getting louder, and they are being pressured to reduce token spend while pressured to release more software.
Some motherfucker today suggested we should use Haiku for all coding tasks, and this guy is a Sr Director of real humans.
these always seem to be written in a vacuum, and certainly not in any mainline silicon valley startup or bigco for the past two years. everyone is stressed and no one is happy.
I think this is a bit misleading, it isn’t laziness.
I played with our AI and it found about 100 real but subtle issues in the codebase. Took less than 30 minutes to find what would have taken a sharp senior months.
I started reviewing the changes manually. They look pretty good but I want to understand each one. Some are design issues and our product needs weren’t spelled out to the AI.
I could blindly put all 100 up for review in half an hour or take a full week to scrutinize them.
So yeah, the process with AI is going to take 100x longer what it could have been.
> Software teams are capturing a lot of the speedup in the form of more leisure or quasi-leisure time. (To greatly simplify: software teams are offering 2x speedups to their companies. They're then using tools that allow them to work 10x as efficiently, and taking nice long coffee breaks.)
That is just an insult. With AI we are working more than ever, either because of "you should be faster now" or because half the people are getting layed-off ...
We write code faster. It's not getting merged faster because we usually find something janky.
Decisions aren't made faster.
Understanding doesn't come faster.
Code doesn't have value. A well working system has value, the bottleneck is now between having code and having a working system, rather than producing the code.
With AI you can write a new Harry Potter novel in under an hour. But if I ask you about the story you can only answer “It involves wizards”.
In pre-AI era you could solve a bug in one week. But you would end up having a detailed mental picture of the system, which requires time to be created. That mental investment helps you make better decisions on the system or solve related problems instantly.
Agreed. The recent Epic Games story (mass layoffs followed by the remaining staff being forced to use AI to crunch) is more in line with reality as I see it. I've been told by management to 'use AI' when I really need to hire.
Who are "we"? We have less (almost none) issues on Sentry, better measured performance, reduced server costs. At the end AI is just writing the same bytecode so you're not going to suddenly see software that didn't exist before.
Coding is only one part of software. There is resourcing, financing, research, planning, design, development, testing, distribution, training and support.
I think that's mainly true. Of course, there are specialized applications necessary to connect the data of one program to another or some fringe cases but by and large new software isn't really needed. A lot of it is coded because of an arms race (hackers vs. security) or to take advantage of people (more ways to make them spend money).
> Software teams are capturing a lot of the speedup in the form of more leisure or quasi-leisure time. (To greatly simplify: software teams are offering 2x speedups to their companies. They're then using tools that allow them to work 10x as efficiently, and taking nice long coffee breaks.)
This might be the biggest factor in my opinion. Corporate incentives usually don't reward someone delivering 2x-10x, so an employee who suddenly found a way to multiply their productivity might be keeping that method to themselves and freeing up some time instead.
AI is a search engine. A good one, better than what was before that. (Which is not saying much, because the search landscape is severely distorted. For example, searching dispropotionately dominates browsing.) It can speed up searching and exploration. This is the natural limit of possible gains.
I thought that for a while, but it does things that go beyond what I can call search. (Search is an elastic term, but not infinitely so.)
For example, it's extraordinarily good at applying the searched-for information (once found) to a specific problem at hand—at least at the first-order level of putting existing pieces together in a correct way. It can do that much faster than I can, with far greater command of detail—seemingly unlimited command of detail, in fact.
It can also find problems and correct them in my own work in a way that still amazes me. Again, not deep problems—they're first-order in the sense of being obvious deviations from correct practice - but to call this 'search' feels like a stretch. It's not searching for bugs in the code I just wrote and then returning a corrected version, as if it had been sitting somewhere waiting to be found.
There are plenty of people saying it can go far beyond just these types of things, and they're probably right, but even these are enough, for me, to refute the it's-just-search view.
The underlying mechanism of that is still search. We do have similar mechanism; we can notice it in dreaming, delirium, "word salad" disorder and so on. It will generate formally "new" things and they will look convincingly normal, especially if our defences are down, as in sleep.
I am not sure what the current state of models is, but as far as I know lifting a limit on the model output length tends to noticeably degrade the quality of the result so that it starts to show the same delirious tones.
Fundamentally it is recalling of past experience, a reaction. In many ways it is an opposite of acquiring true new knowledge. We keep that mechanism in check using other human qualities models do not have.
Working with LLM is basically like hiring a madman. It can do useful work, but you cannot trust it. To be economically viable the verification cycle has to be cheap or the cost of an error must be small. Searching a library or doing a home project are fine. But these aren't the lucrative markets investors are dreaming of.
The software I create for myself has improved a lot.
I also see a lot of improvements in current software and games and fast iterations. But nothing new that is revolutionary.
Almost all of the trending repos on GitHub are _about_ AI: zillion of harnesses and similar AI orchestration tools. So most of the extra efficiency gained by AI usage seems to be spent on building more AI tools and not software for end users or libraries for AIs and humans to build such software.
Developers have never seen a process they didn't fantasize about automating. You see so many of these tools because it's a completely new space, fertile ground, and developers want to automate even more of the development flow so they can spend more time deciding what to build instead of shepherding along a gaggle of agents.
I personally made a bunch of net new, really good software that I use myself and that is much better than alternatives. You will not see that software though, because:
1. Marketing (and distribution) of said software is a chore, and doesn't really give me any benefits
2. It's tailored to me rather than to average Joe.
Same here. I have now the best tools I’ve ever had just because I made them tailored to me and my exact nitpicks.
Advertising them now would just be creating more noise on the internet.
Same here. I'm a bioinformatician, and most of the software I'd have historically been expected to use was cobbled together by biologists then left to rot once the accompanying white paper was published.
I've been able to create my own tooling that beats out the preexisting best across every metric. It's incredible.
There's more of the stuff that's well known and understood. Any software with an ecosystem of integrations and connectors added a lot of connectors lately. CRMs, CMS and similar extended their functionality.
What we've seen very little of is fundamentally new ideas, usage paradigms, user experiences or similar.
It doesn't really matter if we are making better software now, in the end, we know it will be better. Think of it like this: you have a very sick patient, and you prescribe a drug/protocol known to make them worse before they get better. So the intermediate step is not informative in any meaningful sense.
Human coders usually ask this question in malicious way to imply that this technology doesn't just have to be faster than humans (which it is), it also has to produce better code (which it usually does) as cohesive product (so fixing features or refactoring in projects/products do not count) in record time (this is the trap because these models are not even a year old in their current advanced format). It's hopeless.
> Software teams are capturing a lot of the speedup in the form of more leisure or quasi-leisure time. (To greatly simplify: software teams are offering 2x speedups to their companies. They're then using tools that allow them to work 10x as efficiently, and taking nice long coffee breaks.)
I am yet to see the advancements in AI manifest in more leisure-time. It is simply that the amount of work we're expected to be able to pump out has increased, and the actual programming work is no longer what slows us down.
Thanks to AI, I was able to finish my bookkeeping and accounting app I had been working on years ago. https://financica.app
It’s not just “vibe coded”, it’s really damn good. Because thanks to AI, whenever I spot an issue, a quality of life increase etc I can just immediately implement it rather than have it seat in a perpetually growing backlog of “maybe one day”.
And there’s SO MANY standards to comply with especially in Europe and I just … implemented all of them. It’s more full featured than any of its competitors.
It runs the bookkeeping and accounting for all of my own companies plus plenty of customers. And the funny thing is for me, this is pretty much a side project. I just have people helping on sales, but all the development is me.
> I can just immediately implement it rather than have it seat in a perpetually growing backlog of “maybe one day”.
My backlog for my side-project/business had a number of things in it that were frankly never going to happen. They only existed in the backlog so I could say they were in the backlog and add a note about who expressed interest in it so that, if it reached a critical mass, I could justify dedicating the time. That backlog is largely empty now. Things I thought were a pipedream or a "well, maybe one day if this gets bigger" became a reality.
> We are in fact seeing software improvement, in the form of fewer bugs and better performance.
> We are in fact seeing software improvement, in the form of more and better features in existing products.
After reading these two points, I had to scroll down to see if it was the opening of some elaborate joke, because I'm seeing more bugs, worse performance, and less and worse features.
If we released software with the number of defects in the popular coding tools, we'd be out of business before the end of the year. But they persist- for some verticals speed matters over quality.
I built myself a full walkabe 3D plan of my house for my renovation project over the weekend, complete with path tracing renderer, different drawing styles, furniture arrangements and accurate models of different windows styles that I consider ordering, all based on actual blueprints and materials. I looked for usable software for that before but everything was either commercial garbage that wanted you to pony up a lot of money beforehand or hard to use manually. With AI I can build stuff like that over a weekend and I can tell it to place stuff and design things in text and using photos and 3D scans, it takes care of actual implementation. That's much better than spending days or weeks with locked down software trying to make it what I want.
So maybe the bit revolution is that we won't have "big software" that tries to cater to all kinds of users but "small software" that is tailored perfectly to the needs of a single user.
110 comments
[ 0.14 ms ] story [ 19.5 ms ] threadIf the leadership (C-suite) is afraid to take the full responsibility over AI output, then the only rational choice for developers is to "pace the frontier" themselves.
I would like to see some of this software improvement in the form of fewer bugs. If anything it's adding more bugs.
I'll give an example from a game I like called ARC Raiders. For the first ~9 months they did not fix any of the bugs in the main menu. The damn main menu! The first thing you see when you load the game! They also failed to fix, in that time, the bug where simply pressing the Xbox button on the controller would break in-game voice chat.
They only just solved that, as of a few days ago. The main menu bugs still exist.
Anecdotally, completely disagree. Maybe this is nostalgia, but I perceive a distinct and massive drop in software quality over the past few years.
Worst offenders for me are the poster children for LLM-based development. I run into bugs and performance issues with both Claude and Codex literally every day. I don't think I'm an outlier here, either.
AI firm code are likely a different category though.
- more bugs becoming visible that were hidden within the software we use daily
- more bugs (in absolute terms) as more people create more software in a very short amount of time (regardless of their professional training)
- less bugs in existing software as the developers and dev orgs of existing projects is capable of automating bug hunting, test case creation, ... in essence: They get a multiplying effect out of AI regarding their capabilities.
In short: Actual developers with actual experience (the more, the merrier) can do better work while the barrier to enter "the market" for untrained individuals is a lot lower resulting in a huge amount of new software whose "success" is to be determined.
I'm not convinced there are more than a handful of orgs in the world who are disciplined enough to achieve this. Or organisationally willing to increase their QA Eng headcount sufficiently to achieve it.
[1] https://itbranschen.com/wp-content/uploads/2026/05/the-hidde...
Writing code is not the bottleneck anymore, making decisions is. Making the right decisions needs context and thinking. A coffee break might actually help productivity. Driving a Ferrari full speed in a crowded city - only because you can - doesn't make the outcome better.
Even in the 90s, the internet and the web dramatically lowered the costs of distribution, and YC itself was built on top of that (ViaWeb).
Then there has been so many frameworks, libraries, tools, services and APIs to pick and choose from that most of the coding work has been about organization, selling an idea and doing a lot of testing. Which are still unsolved problems.
The only new idea here is the idea that Software Developers are disposable.
This was brought by the same people who did RTO and layoffs post-covid, but now with support of non-tech workers who want a piece of that (supposedly) sweet money.
Always has done. Writing code was never the bottleneck.
Hammock-Driven Development: https://youtu.be/f84n5oFoZBc
Some motherfucker today suggested we should use Haiku for all coding tasks, and this guy is a Sr Director of real humans.
I played with our AI and it found about 100 real but subtle issues in the codebase. Took less than 30 minutes to find what would have taken a sharp senior months.
I started reviewing the changes manually. They look pretty good but I want to understand each one. Some are design issues and our product needs weren’t spelled out to the AI.
I could blindly put all 100 up for review in half an hour or take a full week to scrutinize them.
So yeah, the process with AI is going to take 100x longer what it could have been.
> Software teams are capturing a lot of the speedup in the form of more leisure or quasi-leisure time. (To greatly simplify: software teams are offering 2x speedups to their companies. They're then using tools that allow them to work 10x as efficiently, and taking nice long coffee breaks.)
That is just an insult. With AI we are working more than ever, either because of "you should be faster now" or because half the people are getting layed-off ...
We've lost 75% of the team over the last year.
We write code faster. It's not getting merged faster because we usually find something janky.
Decisions aren't made faster.
Understanding doesn't come faster.
Code doesn't have value. A well working system has value, the bottleneck is now between having code and having a working system, rather than producing the code.
With AI you can write a new Harry Potter novel in under an hour. But if I ask you about the story you can only answer “It involves wizards”.
In pre-AI era you could solve a bug in one week. But you would end up having a detailed mental picture of the system, which requires time to be created. That mental investment helps you make better decisions on the system or solve related problems instantly.
Clear your glasses. For example - https://getartcraft.com/apps Thanks to those people we may soon have great alternatives to monopolies.
Moral compass/integrity stuff. Can't explain why, it just feels like absolutely the wrong direction to be heading in.
Most new tech is just shit.
This might be the biggest factor in my opinion. Corporate incentives usually don't reward someone delivering 2x-10x, so an employee who suddenly found a way to multiply their productivity might be keeping that method to themselves and freeing up some time instead.
For example, it's extraordinarily good at applying the searched-for information (once found) to a specific problem at hand—at least at the first-order level of putting existing pieces together in a correct way. It can do that much faster than I can, with far greater command of detail—seemingly unlimited command of detail, in fact.
It can also find problems and correct them in my own work in a way that still amazes me. Again, not deep problems—they're first-order in the sense of being obvious deviations from correct practice - but to call this 'search' feels like a stretch. It's not searching for bugs in the code I just wrote and then returning a corrected version, as if it had been sitting somewhere waiting to be found.
There are plenty of people saying it can go far beyond just these types of things, and they're probably right, but even these are enough, for me, to refute the it's-just-search view.
I am not sure what the current state of models is, but as far as I know lifting a limit on the model output length tends to noticeably degrade the quality of the result so that it starts to show the same delirious tones.
Fundamentally it is recalling of past experience, a reaction. In many ways it is an opposite of acquiring true new knowledge. We keep that mechanism in check using other human qualities models do not have.
Working with LLM is basically like hiring a madman. It can do useful work, but you cannot trust it. To be economically viable the verification cycle has to be cheap or the cost of an error must be small. Searching a library or doing a home project are fine. But these aren't the lucrative markets investors are dreaming of.
1. Marketing (and distribution) of said software is a chore, and doesn't really give me any benefits 2. It's tailored to me rather than to average Joe.
Wife: Have you got anything without slop?
Waitress: Well, there's slop egg sausage and slop, that's not got much slop in it.
Wife: I don't want ANY slop!
Man: Why can't she have egg bacon slop and sausage?
Wife: THAT'S got slop in it!
Man: Hasn't got as much slop in it as slop egg sausage and slop, has it?
https://news.ycombinator.com/item?id=49539919
I've been able to create my own tooling that beats out the preexisting best across every metric. It's incredible.
What we've seen very little of is fundamentally new ideas, usage paradigms, user experiences or similar.
Human coders usually ask this question in malicious way to imply that this technology doesn't just have to be faster than humans (which it is), it also has to produce better code (which it usually does) as cohesive product (so fixing features or refactoring in projects/products do not count) in record time (this is the trap because these models are not even a year old in their current advanced format). It's hopeless.
I am yet to see the advancements in AI manifest in more leisure-time. It is simply that the amount of work we're expected to be able to pump out has increased, and the actual programming work is no longer what slows us down.
Thanks to AI, I was able to finish my bookkeeping and accounting app I had been working on years ago. https://financica.app
It’s not just “vibe coded”, it’s really damn good. Because thanks to AI, whenever I spot an issue, a quality of life increase etc I can just immediately implement it rather than have it seat in a perpetually growing backlog of “maybe one day”.
And there’s SO MANY standards to comply with especially in Europe and I just … implemented all of them. It’s more full featured than any of its competitors.
It runs the bookkeeping and accounting for all of my own companies plus plenty of customers. And the funny thing is for me, this is pretty much a side project. I just have people helping on sales, but all the development is me.
My backlog for my side-project/business had a number of things in it that were frankly never going to happen. They only existed in the backlog so I could say they were in the backlog and add a note about who expressed interest in it so that, if it reached a critical mass, I could justify dedicating the time. That backlog is largely empty now. Things I thought were a pipedream or a "well, maybe one day if this gets bigger" became a reality.
> We are in fact seeing software improvement, in the form of more and better features in existing products.
After reading these two points, I had to scroll down to see if it was the opening of some elaborate joke, because I'm seeing more bugs, worse performance, and less and worse features.
So maybe the bit revolution is that we won't have "big software" that tries to cater to all kinds of users but "small software" that is tailored perfectly to the needs of a single user.