Author here. Not an anti-AI post. It's about the cognitive cost - faster tasks lead to more tasks, reviewing AI output all day causes decision fatigue, and the tool landscape churns weekly. Wrote about what actually helped. Curious if others are hitting similar walls.
Some of the points raised in the article resonate with me, but I see a lot of trademark phrases inserted by LLMs ("it's not X, it's Y" being the most obvious). Can you share what was your writing process? How much did you write yourself, whether you used LLM to proofread or write the entire text from bullet points, or maybe not at all?
I really feel this. I can make meaningful progress on half a dozen projects in the course of a day now but I end the day exhausted.
I've had conversations with people recently who are losing sleep because they're finding building yet another feature with "just one more prompt" irresistible.
Decades of intuition about sustainable working practices just got disrupted. It's going to take a while and some discipline to find a good new balance.
I loved the section about trying to fight against a system that isn't deterministic.
LLMs because of their nature require constant hand-holding by humans, unless business are willing to make them entirely accountable for the systems/products they produce.
For me the fatigue is a little different— it’s the constant switching between doing a little bit of work/coding/reviewing and then stopping to wait for the llm to generate something.
The waits are unpredictable length, so you never know if you should wait or switch to a new task. So you just do something to kill a little time while the machine thinks.
You never get into a flow state and you feel worn down from this constant vigilance of waiting for background jobs to finish.
I dont feel more productive, I feel like a lazy babysitter that’s just doing enough to keep the kids from hurting themselves
as long as it's new I tremendously enjoy binge watching Claude:
I have three tabs open and if one of them is not doing something interesting I just switch to a different channel, and occasionally influenced the narrative
I've been building https://roborev.io/ (continuous background code review for agents) essentially as a cope to supervise the poor quality of the agents' work, since my agents write much more code than I can possible review directly or QA thoroughly. I think we'll see a bunch of interesting new tools to help alleviate the cognitive burden of supervising their work output.
All these tools are can be a big waste of time if you’re an end user dev. It only makes sense if you are investing your time to eventually use that workflow knowledge to make a product.
1. Make long pauses: 1h of work, stop for 30 minutes or more. The productivity gain should leave you more time to rest. Alternatively work just 50% of time, 2h the morning, 2h the evening, instead of 8 hours. Yet trying to deliver more than before.
2. Don't mix N activities. Work in a very focused way in a single project, doing meaningful progresses.
3. Don't be too open-ended in the changes you do just because you can do it in little time now. Do what really matters.
4. When you are away, put an agent in the right rails to reiterate and provide potentially some very good result in terms of code quality, security, speed, testing, ... This increases the productivity without stressing you. When you return back, inspect the results, discard everything is trash, take the gems, if any.
5. Be minimalistic even if you no longer write the code. Prompt the agent (and your AGENT.md file) to be focused, to don't add useless dependencies, nor complexity, to take the line count low, to accept an improvement only the complexity-cost/gain is adequate.
6. Turn your flow into specification writing. Stop and write your specifications even for a long time, without interruptions. This will improve a lot the output of the coding agents. And it is a moment of calm focused work for you.
On the other side, I feel like using AI tools can reduce the cognitive overload of doing a single task, which can be nice. If you're able to work with a tool that's fast enough and just focus on a single task at a time, it feels like it makes things easier. When you try to parallelize that's when things get messier.
There's a negative for that too - cognitive effort is directly correlated with learning, so it means that your own skills start to feel less sharp too as you do that (as the article mentions)
This write-up has good ideas but gives me the "AI-generated reading fatigue." Things that can cleanly be expressed in 1-2 sentences are whole paragraphs, often with examples that seem unnecessary or unrealistic. There are also some wrong claims like below:
> The Hacker News front page alone is enough to give you whiplash. One day it's "Show HN: Autonomous Research Swarm" and the next it's "Ask HN: How will AI swarms coordinate?" Nobody knows. Everyone's building anyway.
These posts got less than 5 upvotes, they didn't make it to home page. And while overall quality of Show HN might have dropped, HN homepage is still quite sane.
The topic is also not something "nobody talks about," it's being discussed even before agentic tools became available: https://hn.algolia.com/?q=AI+fatigue
I'm also getting really annoyed by AI-generated images like this article has that don't really help comprehension, but make the author feel like they're "pro blogging" because god forbid you have two paragraphs in a row without a subhead or an image.
Programmers complaining about AI but then ripping off umpteen illustrators' labor through AI is infuriating.
Absolute middlebrow dismissal incoming, but the real thinking atrophy is writing blog posts about thinking atrophy caused by LLMs using an LLM.
It is getting very hard to continue viewing HN as a place where I want to come and read content others have written when blog posts written largely with ChatGPT are constantly upvoted to the top.
It's not the co-writing process I have a problem with, it's that ChatGPT can turn a shower thought into a 10 minute essay. This whole post could have been four paragraphs. The introduction was clearly written by an intelligent and skilled human, and then by the second half there's "it's not X, it's Y" reframe slop every second sentence.
The writing is too good to be entirely LLM generated, but the prose is awful enough that I'm confident this was a "paste outline into chatgpt and it generates an essay" workflow.
Frustrating world. I'm lambasting OP, but I want him to write, but actually, and not via a lens that turns every cool thought into marketing sludge.
taking breaks is really something to try and solve in 2026 - to just write regular code, to read, to exercise even. The mind can eventually get overloaded, and there’s no way around proper hygiene.
I only use the free tiers of any particular app. It forces you to really think about you want the tool to do as opposed to treating it as the 'easy' button.
> What should this function be named? I didn't care. Where should this config live? I didn't care. My brain was full. Not from writing code - from judging code.
Does it matter anymore? Most good engineering principles are to ensure code is easy to read and maintain by humans. When we no longer are the target audience for that, many such decisions are no longer relevant.
AI generates a solution that's functional, but that's at a 70% quality level. But then it's really hard to make changes because it feels horrible to spend 1 hour+ to make minor improvements to something that was generated in a minute.
It also feels a lot worse because it would require context switching and really trying to understand the problem and solution at a deeper level rather than a surface level LGTM.
And if it functionally works, then why bother?
Except that it does matter in the long term as technical debt piles up. At a very fast rate too since we're using AI to generate it.
I haven’t hit this yet and now I feel like someone just told me about thorns for the first time while I’m here jogging confidently through the woods with shorts on.
I personally am a lot less stressed. It helped my mood a lot over the last couple of months. Less worries about forgetting things, about missing problems, about getting started, about planning and prioritizing in solo work. Much less of the "swirling mess" feeling. Context switches are simpler, less drudgery, less friction and pulling my hair out for hours banging against some dumb plumbing and gluing issue or installing stuff from github or configuring stuff on the computer.
105 comments
[ 3.0 ms ] story [ 63.3 ms ] threadI've had conversations with people recently who are losing sleep because they're finding building yet another feature with "just one more prompt" irresistible.
Decades of intuition about sustainable working practices just got disrupted. It's going to take a while and some discipline to find a good new balance.
That's the way society is set up.
LLMs because of their nature require constant hand-holding by humans, unless business are willing to make them entirely accountable for the systems/products they produce.
The waits are unpredictable length, so you never know if you should wait or switch to a new task. So you just do something to kill a little time while the machine thinks.
You never get into a flow state and you feel worn down from this constant vigilance of waiting for background jobs to finish.
I dont feel more productive, I feel like a lazy babysitter that’s just doing enough to keep the kids from hurting themselves
2. Don't mix N activities. Work in a very focused way in a single project, doing meaningful progresses.
3. Don't be too open-ended in the changes you do just because you can do it in little time now. Do what really matters.
4. When you are away, put an agent in the right rails to reiterate and provide potentially some very good result in terms of code quality, security, speed, testing, ... This increases the productivity without stressing you. When you return back, inspect the results, discard everything is trash, take the gems, if any.
5. Be minimalistic even if you no longer write the code. Prompt the agent (and your AGENT.md file) to be focused, to don't add useless dependencies, nor complexity, to take the line count low, to accept an improvement only the complexity-cost/gain is adequate.
6. Turn your flow into specification writing. Stop and write your specifications even for a long time, without interruptions. This will improve a lot the output of the coding agents. And it is a moment of calm focused work for you.
On the other side, I feel like using AI tools can reduce the cognitive overload of doing a single task, which can be nice. If you're able to work with a tool that's fast enough and just focus on a single task at a time, it feels like it makes things easier. When you try to parallelize that's when things get messier.
There's a negative for that too - cognitive effort is directly correlated with learning, so it means that your own skills start to feel less sharp too as you do that (as the article mentions)
I'm fatigued by this myth.
> The Hacker News front page alone is enough to give you whiplash. One day it's "Show HN: Autonomous Research Swarm" and the next it's "Ask HN: How will AI swarms coordinate?" Nobody knows. Everyone's building anyway.
These posts got less than 5 upvotes, they didn't make it to home page. And while overall quality of Show HN might have dropped, HN homepage is still quite sane.
The topic is also not something "nobody talks about," it's being discussed even before agentic tools became available: https://hn.algolia.com/?q=AI+fatigue
Programmers complaining about AI but then ripping off umpteen illustrators' labor through AI is infuriating.
"You're not imagining it."
"But my days got harder. Not easier. Harder."
"Now?" as the paragraph opener
"Why? No reason." as the paragraph opener
Nice try OP, submitting your own post to HN.
It is getting very hard to continue viewing HN as a place where I want to come and read content others have written when blog posts written largely with ChatGPT are constantly upvoted to the top.
It's not the co-writing process I have a problem with, it's that ChatGPT can turn a shower thought into a 10 minute essay. This whole post could have been four paragraphs. The introduction was clearly written by an intelligent and skilled human, and then by the second half there's "it's not X, it's Y" reframe slop every second sentence.
The writing is too good to be entirely LLM generated, but the prose is awful enough that I'm confident this was a "paste outline into chatgpt and it generates an essay" workflow.
Frustrating world. I'm lambasting OP, but I want him to write, but actually, and not via a lens that turns every cool thought into marketing sludge.
Does it matter anymore? Most good engineering principles are to ensure code is easy to read and maintain by humans. When we no longer are the target audience for that, many such decisions are no longer relevant.
AI generates a solution that's functional, but that's at a 70% quality level. But then it's really hard to make changes because it feels horrible to spend 1 hour+ to make minor improvements to something that was generated in a minute.
It also feels a lot worse because it would require context switching and really trying to understand the problem and solution at a deeper level rather than a surface level LGTM.
And if it functionally works, then why bother?
Except that it does matter in the long term as technical debt piles up. At a very fast rate too since we're using AI to generate it.
Its a million little quality of life stuff.