Bit of a dunk but ... It must have saved you what ever time you were going to spend on this article because I detect a bunch of LLMisms and very little content.
I'm not sure that Claude saves me time -- I just spent my weekend working on a Claude Code Audio hook with Claude which I obviously wouldn't have worked on elsewise, and that's hardly the gardening I intended to do ... but man it was fun and now my CC sessions are a lot easier to track by ear!
I think a lot of people would disagree with this article on HN, but I’ve yet to see too many people say it’s made their coworkers more productive. That is, do people feel like they’re getting better, more reviewable PRs?
Personally, I’ve been seeing the number of changes for a PR starting to reach into the mid-hundreds now. And fundamentally the developers who make them don’t understand how they work. They often think they do, but then I’ll ask them something about the design and they’ll reply: “IDK Claude did that.”
By no means am I down on AI, but I think proper procedures need to be put into place unless we want a giant bomb in our code base.
AI has allowed me to work on projects I was simply too lazy or haven't attached any importance to. It also allows me to skip the entire process of contacting a designer or doing design work myself (which might actually be a bad thing). The thing I know for sure is that none of my smart home additions I had sitting in a box would be finished if AI didn't exist.
Work harder, create datapoints, democratize knowledge. Except that knowledge will be confined eventually and doom the futures of many people. Use AI now to get ahead of your peers by feeding it questions and evaluating responses. Then in 10 years, Insert Field Here will be dominated by models trained by yesterday’s experts. New members of the field will not be able to compete with the collective knowledge of 1000s of their predecessors. Selling the futures of our youth for short-term gains. It’s quite sad and it is what’s happening.
It’s a shame too because it really could have been something so much more amazing. I’d imagine higher education would shift to how it used to be: a past-time for bored elites. We would probably see a large reduction in the middle class and its eventual destruction. First they went for manufacturing with its strong unions, now they go for the white-collar worker who has little solidarity for his common man (see lack of unions and ethics in our STEM field; most likely because we thought we could never be made redundant). Field by field the middle class will be destroyed and the lower class in thrall of addictive social media, substances, and the illusion of selection into the influencer petty-elite (which remain compliant because they don’t offer value proportional to the bribes they receive). The elites will have recreated the dynamic that existed for most of human history. Final point, see the obsession of current elites in using artificial insemination to create a reliable and durable contingent of heirs. Something previous rulers could only dream about in history.
The blogosphere (am I dating myself?) keeps bringing up the METR study (https://metr.org/blog/2025-07-10-early-2025-ai-experienced-o...) without really understanding the result. The guy with experience had a huge boost. You are reading the results wrong if your conclusion is this blog.
Here's the kicker: AI was supposed to automate the boring parts so we could “focus on high-leverage, strategic, needle-moving, synergistic core competencies.” Instead, we’re stuck in a recursive loop of prompt engineering, hallucination triage, output validation, re-prompting, Slack channel FOMO, and productivity theater. We’ve basically replaced “doing the work” with “managing the tool that kinda tries to do the work but needs babysitting.” Congrats—we’ve invented Jira for thought. And here's the kicker.
I've created hundreds of small scripts that I wouldn't have bothered with before and either do some manual checks or just not possess the information. Just on "small script" productivity it already saved me a lot of time.
The problem is people trying to make the models do things that are too close to their limit. You should use LLMs for things they can ace already, not waste time trying to get it to invent some new algorithm. If I don't 0-3 shot a problem then I will just either do it manually or not do it.
Similarly to giving up on a Google search that you try a few times and nothing useful comes in the first few prompts. You don't keep at it the whole afternoon.
> But here’s the kicker: we were told AI would free us up for “higher-level work.” What actually happened? We just found more work to fill the space. That two-hour block AI created by automating your morning reports? It’s now packed with three new meetings. The 30 minutes you saved on data analysis? You’re using it to manage two more AI tools and review their outputs.
This is basically the definition of increased productivity and efficiency. Doing more stuff in the same amount of time. What I tell people who are anxious about whether their job might be automated away by AI is this:
We will never run out of the problems that need solving. The types of problems you spend your time solving will change. The key is to adapt your process to allocate your time to solving the right kinds of problems. You don’t want the be the person taking an hour to do arithmetic by hand, when you have access to spreadsheets.
AI efficiency gains don’t benefit employees, they benefit _employers_, who get more output from the same salary.
When you’re salaried, you’re selling 8 hours of time, not units of work. AI that makes you 20% faster doesn’t mean you work 20% fewer hours or get a 20% raise. It means your employer gets 20% more value from the same labor cost.
Marx: workers sell their capacity to work for a fixed period, and any productivity improvements within that time become surplus value captured by capital.
AI tools are just the latest mechanism for extracting more output from the same wage.
The real issue isn’t the technology—it’s that employees can’t capture gains from their own efficiency improvements. Until compensation models shift from time-based to outcome-based, every productivity breakthrough just makes us more profitable to employ, not more prosperous ourselves.
It’s the Industrial Revolution all over again and we’re the Luddites
Come on, what did you think was going to happen? The historical record has consistently shown that humans have not worked less when given tools that increased productivity; they simply produced more.
Economist Keynes predicted one century ago that the workweek would drop to 15 hours due to rising productivity. It has not happened for social reasons.
I don't know what's going to happen when humans become redundant; that's an incipient issue we'll have to grapple with.
AI is bringing efficiency, and it is making us work harder. It’s because AI marginally improves certain workflows but the management is using that to fire employees and offload their work on the remaining ones. You get 1.2x improvement in efficiency but get 2x the work.
It was ever thus. In each technological revolution—from the Industrial Revolution to personal computing revolution—the promise is that the technology will make work so efficient and productive that we can all work less.
Unfortunately, it is always a deliberate lie by the people who stand to gain from the new technology. Anyone who has thought about it for five seconds knows that this is not how capitalism works. Productivity gains are almost immediately absorbed and become the new normal. Firms that operate at the old level of productivity get washed out.
I simply can't believe that we're still falling for this. But let's hold out hope. Maybe AGI is just around the corner, and literally everyone in the world will spend our time sipping margaritas on the beach while we count our UBI. Certainly AI could never accelerate wealth concentration and inequality, right? RIGHT?
At least AI gave the author an easy think piece without having to write anything in their own style.
"But here's the kicker"
"It's not x. It's y."
"The companies that foo? They bar."
Em-dashes galore.
I'm either hypersensitized, seeing ghosts, or this article got the "yo claude make it pop" treatment. It's sad, but anything overly polished immediately triggers some "is this slop or an original thought actually worth my time" response.
I don't get these articles. First the author claims that AI made us more productive but now the time we saved is spent on more work (!)
> But here’s the kicker: we were told AI would free us up for “higher-level work.” What actually happened? We just found more work to fill the space. That two-hour block AI created by automating your morning reports? It’s now packed with three new meetings.
But then a few sentences later, she argues that tools made us less productive.
> when developers used AI tools, they took 19% longer to complete tasks than without AI. Even more telling: the developers estimated they were 20% faster with AI—they were completely wrong about their own productivity.
Then she switches back to saying it saves us time but cognitive debt!
> If an AI tool saves you 30 minutes but leaves you mentally drained and second-guessing everything, that’s not productivity—that’s cognitive debt.
I think you have to pick one, or just admit you don't like AI because it makes you feel icky for whatever reason so you're going to throw every type of argument you can against it.
This isn't surprising to me. My experience is that AI is best suited for a single person to get started rapidly with a new project or, if used artfully, to quickly orchestrate refactoring that the user has planned. I've personally found AI to be good for my productivity (haven't measured, however, my work is not conducive to that kind of thing) but I've also found I use AI primarily to look up documentation and to type code out. I still think about software design as much as I ever have, whether its the initial design step or refactoring.
We've had technological progress that rapidly shifts the number of person-hours per <output> for generations. We don't have to guess. We've seen this play out many times already.
At first, we spend our time one way (say eight hours, just to pick a number). Then we get the tools to do all of that in six hours. Then when job seeking and hiring, we get one worker willing to work six hours and another willing to work eight, so the eight-hour worker gets the job, all else equal. Labor is a marketplace, so we work as much as we're willing to in aggregate, which is roughly constant over time, so efficiency will never free up individuals' time.
In the context of TFA, it means we just shift our time to "harder" work (in the sense of work that AI can't do yet).
Only the the context of weak labor laws and no unions.
There are more workers than CEOs and CEOs don't create the value, the workers do.
What CEOs largely do is monetize and take credit for all the value creation.
I don't think things can continue along the current unsustainable path in the US. Something will break; hopefully in a sensible way and not the historical pitchforks and arson.
> Then when job seeking and hiring, we get one worker willing to work six hours and another willing to work eight, so the eight-hour worker gets the job, all else equal.
Do you think people get hired by bidding for number of hours they want to work?
My experience with AI coding is it might be slower to develop for short term, but it’s saving ton of time for the long term.
Here is an example.
I decided to create a new app, so I write down a brief of what it should do, ask AI to create a longer readme file about the platform along with design, sequence diagram, and suggested technologies.
I review that document, see if there is anything I can amend, then ask AI for the implementation plan.
Up until this point, this would probably increased the time I usually use to describe the platform in writing. But realistically, designing and thinking about systems were never that fast. I would have to think about use cases, imagine workflows in my mind, do pen and paper diagrams which I don’t think any of the productivity reports are covering.
In my team, it's making one dev more prolific, and everybody else work harder.
The most junior dev on my team was tasked with setting up a repo for a new service. The service is not due for many many months so this was an opportunity to learn. What we got was a giant PR with hundreds of new configurations no one has heard of. It's not bad, it's just that we don't know what each conf does. Naturally we asked him to explain or give an overview, he couldn't. Well because he fed the whole thing to an LLM and it spat out the repo. He even had fixes for bugs we didn't know we had in other repos. He didn't know either. But it took the rest of the team digging in to figure out what's going on.
I'm not against using LLM, but now I've added a new step in the process. If anyone makes giant PRs, they'll also have to make a presentation to give everyone an overview. With that in mind, it forces devs to actually read through the code they generate and understand it.
you will never be given your time back by an employer. you have to take it. you might be able to ask for it, but it won't be freely given, whether or not you become more efficient. LLM chatbots and agents are, in this sense, just another tool that changes our relationship to the work we do (but never our relationship to work).
Just like with "automatic checkout systems" at a grocery store. Passing the labor onto the individual, vs. the expert. We don't even get the same infrastructure professionals get. A PLU for a piece of fruit is a mind-blurring whisk of hands over a dial pad for a pro, and a bit of mind-numbingly arduously piss-poor series of taps for the ill-positioned entrant.
35 comments
[ 2.3 ms ] story [ 46.9 ms ] threadI'm not sure that Claude saves me time -- I just spent my weekend working on a Claude Code Audio hook with Claude which I obviously wouldn't have worked on elsewise, and that's hardly the gardening I intended to do ... but man it was fun and now my CC sessions are a lot easier to track by ear!
Personally, I’ve been seeing the number of changes for a PR starting to reach into the mid-hundreds now. And fundamentally the developers who make them don’t understand how they work. They often think they do, but then I’ll ask them something about the design and they’ll reply: “IDK Claude did that.”
By no means am I down on AI, but I think proper procedures need to be put into place unless we want a giant bomb in our code base.
Edit: "amirhirsch" user probably explained this better than me in an above comment.
It’s a shame too because it really could have been something so much more amazing. I’d imagine higher education would shift to how it used to be: a past-time for bored elites. We would probably see a large reduction in the middle class and its eventual destruction. First they went for manufacturing with its strong unions, now they go for the white-collar worker who has little solidarity for his common man (see lack of unions and ethics in our STEM field; most likely because we thought we could never be made redundant). Field by field the middle class will be destroyed and the lower class in thrall of addictive social media, substances, and the illusion of selection into the influencer petty-elite (which remain compliant because they don’t offer value proportional to the bribes they receive). The elites will have recreated the dynamic that existed for most of human history. Final point, see the obsession of current elites in using artificial insemination to create a reliable and durable contingent of heirs. Something previous rulers could only dream about in history.
It disgusts me and pisses me off so much.
And that was before Claude Code.
The problem is people trying to make the models do things that are too close to their limit. You should use LLMs for things they can ace already, not waste time trying to get it to invent some new algorithm. If I don't 0-3 shot a problem then I will just either do it manually or not do it.
Similarly to giving up on a Google search that you try a few times and nothing useful comes in the first few prompts. You don't keep at it the whole afternoon.
This is basically the definition of increased productivity and efficiency. Doing more stuff in the same amount of time. What I tell people who are anxious about whether their job might be automated away by AI is this:
We will never run out of the problems that need solving. The types of problems you spend your time solving will change. The key is to adapt your process to allocate your time to solving the right kinds of problems. You don’t want the be the person taking an hour to do arithmetic by hand, when you have access to spreadsheets.
Marx: workers sell their capacity to work for a fixed period, and any productivity improvements within that time become surplus value captured by capital.
AI tools are just the latest mechanism for extracting more output from the same wage. The real issue isn’t the technology—it’s that employees can’t capture gains from their own efficiency improvements. Until compensation models shift from time-based to outcome-based, every productivity breakthrough just makes us more profitable to employ, not more prosperous ourselves.
It’s the Industrial Revolution all over again and we’re the Luddites
This is what the whole four-day workweek movement is about; to reclaim some of that productivity increase as personal time. https://en.wikipedia.org/wiki/Four-day_workweek
Economist Keynes predicted one century ago that the workweek would drop to 15 hours due to rising productivity. It has not happened for social reasons.
I don't know what's going to happen when humans become redundant; that's an incipient issue we'll have to grapple with.
Unfortunately, it is always a deliberate lie by the people who stand to gain from the new technology. Anyone who has thought about it for five seconds knows that this is not how capitalism works. Productivity gains are almost immediately absorbed and become the new normal. Firms that operate at the old level of productivity get washed out.
I simply can't believe that we're still falling for this. But let's hold out hope. Maybe AGI is just around the corner, and literally everyone in the world will spend our time sipping margaritas on the beach while we count our UBI. Certainly AI could never accelerate wealth concentration and inequality, right? RIGHT?
"But here's the kicker"
"It's not x. It's y."
"The companies that foo? They bar."
Em-dashes galore.
I'm either hypersensitized, seeing ghosts, or this article got the "yo claude make it pop" treatment. It's sad, but anything overly polished immediately triggers some "is this slop or an original thought actually worth my time" response.
> But here’s the kicker: we were told AI would free us up for “higher-level work.” What actually happened? We just found more work to fill the space. That two-hour block AI created by automating your morning reports? It’s now packed with three new meetings.
But then a few sentences later, she argues that tools made us less productive.
> when developers used AI tools, they took 19% longer to complete tasks than without AI. Even more telling: the developers estimated they were 20% faster with AI—they were completely wrong about their own productivity.
Then she switches back to saying it saves us time but cognitive debt!
> If an AI tool saves you 30 minutes but leaves you mentally drained and second-guessing everything, that’s not productivity—that’s cognitive debt.
I think you have to pick one, or just admit you don't like AI because it makes you feel icky for whatever reason so you're going to throw every type of argument you can against it.
At first, we spend our time one way (say eight hours, just to pick a number). Then we get the tools to do all of that in six hours. Then when job seeking and hiring, we get one worker willing to work six hours and another willing to work eight, so the eight-hour worker gets the job, all else equal. Labor is a marketplace, so we work as much as we're willing to in aggregate, which is roughly constant over time, so efficiency will never free up individuals' time.
In the context of TFA, it means we just shift our time to "harder" work (in the sense of work that AI can't do yet).
There are more workers than CEOs and CEOs don't create the value, the workers do. What CEOs largely do is monetize and take credit for all the value creation.
I don't think things can continue along the current unsustainable path in the US. Something will break; hopefully in a sensible way and not the historical pitchforks and arson.
Do you think people get hired by bidding for number of hours they want to work?
Here is an example.
I decided to create a new app, so I write down a brief of what it should do, ask AI to create a longer readme file about the platform along with design, sequence diagram, and suggested technologies.
I review that document, see if there is anything I can amend, then ask AI for the implementation plan.
Up until this point, this would probably increased the time I usually use to describe the platform in writing. But realistically, designing and thinking about systems were never that fast. I would have to think about use cases, imagine workflows in my mind, do pen and paper diagrams which I don’t think any of the productivity reports are covering.
The most junior dev on my team was tasked with setting up a repo for a new service. The service is not due for many many months so this was an opportunity to learn. What we got was a giant PR with hundreds of new configurations no one has heard of. It's not bad, it's just that we don't know what each conf does. Naturally we asked him to explain or give an overview, he couldn't. Well because he fed the whole thing to an LLM and it spat out the repo. He even had fixes for bugs we didn't know we had in other repos. He didn't know either. But it took the rest of the team digging in to figure out what's going on.
I'm not against using LLM, but now I've added a new step in the process. If anyone makes giant PRs, they'll also have to make a presentation to give everyone an overview. With that in mind, it forces devs to actually read through the code they generate and understand it.
you will never be given your time back by an employer. you have to take it. you might be able to ask for it, but it won't be freely given, whether or not you become more efficient. LLM chatbots and agents are, in this sense, just another tool that changes our relationship to the work we do (but never our relationship to work).