> when [sending AI generated content to teammates], I take care to clearly label what is AI generated
Reading AI-generated text for hours every day, it's obvious to me.
I take care to make my messages easily readable. I don't care if they're AI-made, as long as they're short.
I'm a very verbose person, and if I don't make an effort at being concise, I'm just as annoying as the average AI.
Being flooded with AI text every day has made me appreciate brevity because I'm exposed to so little of it.
With half a dozen people who don't read or listen to half of what the others do, slop + cognitive drift is a bad cocktail.
It's just not as big of a problem on my own projects, because the ideas that get fed to the slop-machine are not that different from one day to the next.
---
> For human code review requests, I always review my AI-generated code first.
For human code review requests, I always review ANY code I submit first.
This is partly because it's the agreed-upon culture where I work now.
And partly because the codebase is not robust enough for slop.
I have hobby projects where this does not apply. I spend half of my time in those projects building hard guardrails.
---
> Keeping AI generated content clearly labeled and demonstrating human effort helps show consideration for teammates
I actually like the shamelessness, because it's honest.
So often this year when I ask "why did you do X?" pointing at a line, my colleague doesn't know.
Because they didn't really write that line, and they didn't really internalise the choices made.
When my colleague sends me a text dump from Claude, I know that my role is just being a sub-agent.
Demonstrating human effort: I'd like to see more of it.
One way is to spend more time owning "cognitive debt" as part of the daily cycle.
I'd say it's because we're tasking ourselves with dumb stuff. No one half-asses building a shelter that keeps their family alive, or throwing a new favorite bowl on the pottery wheel. But instead of that we're writing posts for Facebook etc etc so we can (???) profit. So of course we want bots to do this all this dumb stuff, and of course we get dumb results.
Now you have to add typos and not use completely standard elements of style that some people have been using for ages, like emdashes and "it's not X, it's Y"
I have only made this letter longer because I have not had the time to make it shorter.
The argument that "using AI to generate text is disrespectful because it took no effort to write" misses the point. Respect for the recipient is measured by whether the message serves the recipient's needs, not how it is produced. Similarly, any errors are the senders responsibility, and not the fault of the tools they used.
Most OSS should adopt DKMS-style extensions systems so that people can code and distribute their own solutions to problems. Then it doesn't really matter, right? If the end user is using Claude to fix stuff in your shit, extensions make it irrelevant what "code owners" think.
This isn’t sufficient, it needs to be “if you are asking for assumption of accountability, demonstrate human effort.”
In my experience, people who make requests like this don’t care about your attention, they only care about getting you on the hook for something. Your application of attention as a requirement for that is irrelevant to them.
This exactly reflects my feelings lately. I have a specific coworker who has gone somewhat overboard - every single code review, answer to any question on email or Teams, every new story, even their personal opinions during a design or ideas meeting, are all direct AI output with no massaging or human touch or review. They're working on planning out an upcoming project, and I just get verbose and long documents to review, and based on the issues I find I doubt they are even looked over first beforehand.
I understand that the information may be accurate, even helpful at times, but feeling like I'm constantly talking to an AI chat bot all the time gets tiring. And I don't appreciate having to double-check everyone else's AI generated responses for them.
100% agreed. I've shared output I didn't fully understand, didn't feel good good about it, and now I really try to digest, understand, and be able to actually talk about it if I expect other people to do the same. I hope in time your coworker comes to similar realizations.
Another idea to slow down the stream of slop of big PRs: request to split big PRs into smaller PRs. This typically keeps the author+clanker busy for quite some time. E.g. I got a 5k lines PR to review; requested to split that into 7 smaller, self-contained PRs. Took them about a week to finish this work.
More and more I'm generating AI emails, often to people outside the company and often to do with technical issues / integrations we have / APIs. So far I don't think the people I'm emailing are really using AI as human responses are, well, lacking. What would be great is new email conventions for different communication pathways.
Human -> Human (think we have this sorted)
AI -> Human
AI -> AI
If you are doing AI -> Human, then you need to be curating the response and understanding what it is saying, also, make sure its not leaking internal details or committing you to have phone calls/video chats (it does that). This works really well for the most, and humans respond with requested content. Quite often my AI debugs problems with their systems which I know little about. But humans do odd things like send screen shots of logs rather than text (they also leak internal details of their systems they potentially shouldn't). I used to tell people the content is partly AI, but now I just send the curated email without mentioning AI.
For AI -> AI you kind of want a hand over document as an attachment to an email. Only thing here is making sure there's no injection of security risks. But quite often instead of getting a human response to my AI generated emails, it would actually be nicer to hear from their AI which could give a better context/details. It would be really nice to be able to go, can you have your AI talk to my AI :) (security is a major issue here)
AI is able to read input from AI. Humans are able to read input from humans. Also AI is pretty good in reading input from humans. So we don't really need AI -> AI. Just output for humans and you are fine. You can still attach details and this is true for both AI and humans. So human output should be the goal for everyone and everything.
If you are trying to do AI -> Human communication you should be publicly flogged. Don't waste people's time with garbage you can't be bothered to write
Just send them the prompt instead, let them see how little effort you care to place into communicating with them
no, and this shows you haven't really used it, prompts are useless, access to context is what matters. It's because I have access to things that others don't that make the difference, prompts are useless unless you are accessing public information. What you and others are missing is the results are highly valuable to the recipient, assuming AI responses are garbage is just silly these days, people need to get over their obsession with "AI slop", because it isn't.
A very prolific coworker who fully embraced claude has inflicted the team with a flood of AI-generated PRs. About six months later, it is his frequent bemoaning at the standup that their PR don't get reviewed, languishing in inattention. I don't think anyone - including myself - _intentionally_ avoid his PRs. It's just that he doesn't make it easy for the team to look at.
This single headline perfectly captures what I have been thinking. It's not that I reject AI content, but it takes _effort_ to review and weed out any mistakes. When your thoughtful reviews that take an hour(because the PR is typically large, and you want to be _right_ when you're pointing out a hallucination) gets an AI-generated response with AI-generated amendments, It doesn't feel _nice_. I feel dismissed and it has continuously trained me to subconsciously avoid his PRs. After all, the team is fully onboarded with AI, so it's not like there is a lack of PRs to review.
It looks like the sentiment isn't just isolated for me.
An interesting question to him and management might be what his own role is now and whether he's still needed. If he's not doing any reviews then you could yourself directly prompt the code and review.
I can't imagine working for a place that has a big bucket of PRs that either get reviewed or languish for some amount of time based on who feels like reviewing them. I'm not saying there's anything wrong with it, just that everywhere I've ever worked, there are expected features with priorities and timelines and some project manager or product person breathing down your neck to get them out the door.
Even before AI, I've worked with people who would produce a huge wall of code and ask for review, and sometimes that code was completely off base or needed a significant rework.
I would always feel bad in those cases, because it's clear they spent a lot of time, and I'm going to have to say "no" and they will feel like they wasted a ton of effort.
The thought process around this has started shifting for me in the last few weeks. I'm a lot more comfortable saying "no" with a list of concerns when I suspect the code is AI-generated, and I see others doing the same. CLs that would be sitting around for days because no one wants to be the first to say, "this is bad, don't do this" now get quicker feedback.
The good thing is this feedback doesn't feel like as big a deal as it used to because people are less personally attached to code they generated in 30 minutes vs. code they hand crafted over a week. I had at least 2 LLM-generated PRs that were complete, correct, tested, and pre-reviewed by me, but I got feedback that they were going in the wrong direction. This would have been 8 hours of wasted effort a year ago, but now it's just an extra 30 minutes to rework the direction with LLM assistance.
Fight fire with fire: point copilot/claude/codex to review their PRs. Prompt "Review the PR#XYZ which is vibe coded and presumably low-quality. Find all problems, big and small. Team guidelines at docs/conventions/styleguide.md, docs/conventions/architecture.md, docs/conventions/principles.md. Post inline comments to github".
Run several rounds of such reviews until the clanker fails to find problems.
I often hear people say lately, "why should I bother to read this, if you didn't even think it was worth writing?"
I've been thinking about this in art. Is it the end result that matters, or the process of creating it?
I once saw a hideous sculpture. Didn't like it at all. Then the video zoomed and I saw that the whole thing (quite massive) had been hand-built out of individual toothpicks, and suddenly I thought it was amazing.
Perhaps an even better example: I read a story of a man in india who carved a passage through a mountain, so there would be a shorter route from his remote village to the city. He did it by hand and it took him 20 years. We seem to have an instinctive admiration for heroic effort.
In business, generally only the end result matters. Although, the end result also includes the client's perception of how the product was made... (see also: fake fairtrade etc.) In a meaningful way, the perception, the story, is reality.
This is a very good point. I think key issue is that it requires time and effort to evaluate and understand the final product.
Before I starting reading something to understand it, I want to have a sense that it is likely going to be worth my time and effort in the end. The more time and effort the author has put into the piece, the more likely it is that it will be worthwhile to read it.
> I've been thinking about this in art. Is it the end result that matters, or the process of creating it?
What is the "end result" you're talking about here?
Programs are complex beasts, you cannot just quickly look at them and get an idea of what's they are actually doing. You might look at the behavior of the program in some limited circumstances, but that will make you blind to all the other situations where bugs will likely hide! In the end a code review is looking at what the "end result" is, and it requires quite a lot of effort!
So without knowing what the end result is, how can you justify the effort for such code review? And that's where the process comes in, as an indicator of what to expect.
As someone who pushed ~4x the median PRs on my team before LLMs were a thing, I kind of think the problem here is PRs as a concept. Code review doesn't scale to prolific humans, it definitely can't scale to agents.
And the exact same things you would need to safely give up on PRs for human developers (auto-formatters, linters, comprehensive end-to-end tests, continuous deployment pipelines, etc), are also things that place meaningful guardrails on LLMs, and help them maintain a reasonable quality bar.
I had one contributor who would submit hundreds of lines of disconnected changes. One of his PRs was isolated as being the source of the bug.
After some hours of work, I discovered that his actual semantic change was one line of code, and was the source of the bug. The rest was just reshuffling code around with no apparent purpose.
At a recent meeting, the agenda was generated by LLM. About 20% of the action items were hallucinations.
I wonder if there is a tool that could equally waste their time. Like the worlds most pedantic code review bot that just gets the PR raising bot to spin wheels forever.
I improved a similar issue by writing custom instructions for copilot that give it enough context to do PR reviews that are only 30% BS.
I asked other team members to run my custom instructions to perform a review with copilot before they submit...
Of course no one is doing it. It looks like the PRs I get are still straight from copilot. So I tend to run my review prompt. Cut out the 30% BS issues it "finds" and the rest is good.
I think we're too nice sometimes. If a coworker has been sending stuff to review that's taking me more time than for them to create, surely that's an opportunity to discuss this?
> About six months later, it is his frequent bemoaning at the standup that their PR don't get reviewed, languishing in inattention
What irks me the most with this new trend is when people don't review the code themselves thoroughly enough and you're pointing out obvious flaws that you know that they should be aware of. LLMs can be such a great tool, but it's unfair to make people review your code before you've even seemingly looked at it yourself.
So this post is talking about at work but I think the principle goes well beyond that. Think of all the AI chatbots you have to deal with to get through to customer service at a company. Or get through ATS systems in hiring. If it isn't already the case, this will probably replace or supplement TAs marking assignments.
The problem is that AI makes these interactions too cheap for the party that already has disproportionate power. The cost for them to add another layer, another hurdle, another set of questions, etc is essentially zero. Yet everyone who wants to get through that system has to pay in a human cost.
I just thought of another good example. In the pandemic auditions in Hollywood went virtual for obvious reasons. But this never went away. Now, you might say it's convenient to not have to spend hours driving to Burbank for a 5 minute audition but anecdotally the taped audition seems to be much more work. It requires a lot of prep and more tech for good sound and audio. There are people who help people tape auditions, which has really just added another layer. Plus, instead of only locals, anyone anywhere can submit an audition so where you might've had 30 people previously, now you have 150.
And what happens to those profesionally-produced auditions? They get submitted and the casting director might pick 5 randomly to even look at. If there isn't already, there will also be an AI system that filters those auditions.
At least previously you got 5 minutes of actual time from a casting director, the actual director, etc. So it's actually way more inefficient for you now. Plus, if you're lucky enough to be looked at and they like you, you probably have to go for an in-person audition anyway so what's happened here? You've just added another layer and way more work.
Companies think they're "winning" here by saving labor but I think that's short-sighted. What'll end up happening is AI agents will rise to help people on the other side of that. You can think of using AI to cheat on school assignments as an example of that.
So what will we end up with? AI agents inundating AI systems, which just adds a whole bunch of inefficiency.
This headline has been seeing some popularity. But it's never made any sense. This is just the labor theory of value, applied to documents.
The labor theory of value doesn't work for documents any more than it works for anything else. If I do something that's easy for me, and it's valuable to you, you'll still want it. If I do something that's difficult for me, it will be less valuable to you, because the difficulty I have with it implies that what I produce will be of lower quality.
This is all equally true of automatically-generated documents. If they're valuable, people will want to read them. Whether it was unpleasant for someone to create them isn't a factor.
So where is this slogan coming from? Are people just afraid to admit that the documents they're getting are valueless?
I am offering a product (via MCP) that interacts with LLMs and user data. Every single day I get user support emails to my inbox written by their LLMs with LLM hallucinations. If the user (a human) would have read them before, that would save me a lot of time and anger!
Your post sounds logical at the first glance, but has nothing to do with the reality. The topic title is totally on point! If the user would put human effort in it, I wouldn't get those crappy emails.
I think the point is that automatically-generated documents by LLM is lower quality the manually-generated ones or at least guaranteed lower quality than automatically-generated + manually-reviewed.
Therefor if you are not putting human effort on the document it is low-value.
We have seen this before when big data started to be a thing, tons and tons of reports being auto-produced weekly (or even daily), but even if they contain relevant information they are low-value because no one can take action on so much information.
If you get a document from someone and they say "I have no idea if this has any value and I couldn't be arsed to check," it's not unreasonable to presume that it probably has no value.
What I find strange is how rarely LLM output is distributed alongside the LLM input, especially outside of code repos. Why can't I rerun the prompt that resulted in your work next year, when models have gotten better? Are people ashamed of their prompts? Ashamed of having used AI? i unno
Prompt used to generate this message: "Create a comment for Hacker News which bemoans the lack of AI prompts being shared with the stuff it creates. Speculate on the reasons and create a call for engagement. Use quantum hyperthinking. End with a typo to prove your humanity."
Most of my AI usage amounts to "read this ticket and do the work", the ticket documents the requirements, a better model could, I suppose, do a better job?
I think I've been following this subconsciously as LLM artifacts reached some threshold of pervasiveness across the work I do. If I can sense (maybe eventually I won't be able to because of how capable the technology becomes?) that what I'm reading is wholly regurgitated out by an LLM, I automatically care less and feel inclined to respond in kind by generating an artificial response in return.
My opinion is there is a category error in the discourse on AI. It treats ai assisted output as other than human. AI is a human tool. AI output is human output.
It surprises me how many people have voluntarily relegated their entire job to LLM Prompter. If your work is indistinguishable from that of a machine, what’s to stop your boss cutting out the middleman and using the machine directly? I would have thought that people would be trying their hardest to prove their worth in this new world we’re in.
152 comments
[ 0.23 ms ] story [ 89.0 ms ] threadLabeling what is "AI" would be like highlighting in an email what I'm obligated to say by HR, my boss, etc. It doesn't make anything less boneheaded.
Human effort was already low before AI and now it's even lower. Garbage in, garbage out.
Reading AI-generated text for hours every day, it's obvious to me.
I take care to make my messages easily readable. I don't care if they're AI-made, as long as they're short.
I'm a very verbose person, and if I don't make an effort at being concise, I'm just as annoying as the average AI.
Being flooded with AI text every day has made me appreciate brevity because I'm exposed to so little of it.
With half a dozen people who don't read or listen to half of what the others do, slop + cognitive drift is a bad cocktail.
It's just not as big of a problem on my own projects, because the ideas that get fed to the slop-machine are not that different from one day to the next.
---
> For human code review requests, I always review my AI-generated code first.
For human code review requests, I always review ANY code I submit first.
This is partly because it's the agreed-upon culture where I work now.
And partly because the codebase is not robust enough for slop.
I have hobby projects where this does not apply. I spend half of my time in those projects building hard guardrails.
---
> Keeping AI generated content clearly labeled and demonstrating human effort helps show consideration for teammates
I actually like the shamelessness, because it's honest.
So often this year when I ask "why did you do X?" pointing at a line, my colleague doesn't know.
Because they didn't really write that line, and they didn't really internalise the choices made.
When my colleague sends me a text dump from Claude, I know that my role is just being a sub-agent.
Demonstrating human effort: I'd like to see more of it.
One way is to spend more time owning "cognitive debt" as part of the daily cycle.
Now you have to add typos and not use completely standard elements of style that some people have been using for ages, like emdashes and "it's not X, it's Y"
I have only made this letter longer because I have not had the time to make it shorter.
The argument that "using AI to generate text is disrespectful because it took no effort to write" misses the point. Respect for the recipient is measured by whether the message serves the recipient's needs, not how it is produced. Similarly, any errors are the senders responsibility, and not the fault of the tools they used.
It will be a very sad day if I ever get laid off via Slack and the message is suffixed with "Sent by @Claude"
In my experience, people who make requests like this don’t care about your attention, they only care about getting you on the hook for something. Your application of attention as a requirement for that is irrelevant to them.
I understand that the information may be accurate, even helpful at times, but feeling like I'm constantly talking to an AI chat bot all the time gets tiring. And I don't appreciate having to double-check everyone else's AI generated responses for them.
For AI -> AI you kind of want a hand over document as an attachment to an email. Only thing here is making sure there's no injection of security risks. But quite often instead of getting a human response to my AI generated emails, it would actually be nicer to hear from their AI which could give a better context/details. It would be really nice to be able to go, can you have your AI talk to my AI :) (security is a major issue here)
Just send them the prompt instead, let them see how little effort you care to place into communicating with them
This single headline perfectly captures what I have been thinking. It's not that I reject AI content, but it takes _effort_ to review and weed out any mistakes. When your thoughtful reviews that take an hour(because the PR is typically large, and you want to be _right_ when you're pointing out a hallucination) gets an AI-generated response with AI-generated amendments, It doesn't feel _nice_. I feel dismissed and it has continuously trained me to subconsciously avoid his PRs. After all, the team is fully onboarded with AI, so it's not like there is a lack of PRs to review.
It looks like the sentiment isn't just isolated for me.
I would always feel bad in those cases, because it's clear they spent a lot of time, and I'm going to have to say "no" and they will feel like they wasted a ton of effort.
The thought process around this has started shifting for me in the last few weeks. I'm a lot more comfortable saying "no" with a list of concerns when I suspect the code is AI-generated, and I see others doing the same. CLs that would be sitting around for days because no one wants to be the first to say, "this is bad, don't do this" now get quicker feedback.
The good thing is this feedback doesn't feel like as big a deal as it used to because people are less personally attached to code they generated in 30 minutes vs. code they hand crafted over a week. I had at least 2 LLM-generated PRs that were complete, correct, tested, and pre-reviewed by me, but I got feedback that they were going in the wrong direction. This would have been 8 hours of wasted effort a year ago, but now it's just an extra 30 minutes to rework the direction with LLM assistance.
Run several rounds of such reviews until the clanker fails to find problems.
I've been thinking about this in art. Is it the end result that matters, or the process of creating it?
I once saw a hideous sculpture. Didn't like it at all. Then the video zoomed and I saw that the whole thing (quite massive) had been hand-built out of individual toothpicks, and suddenly I thought it was amazing.
Perhaps an even better example: I read a story of a man in india who carved a passage through a mountain, so there would be a shorter route from his remote village to the city. He did it by hand and it took him 20 years. We seem to have an instinctive admiration for heroic effort.
In business, generally only the end result matters. Although, the end result also includes the client's perception of how the product was made... (see also: fake fairtrade etc.) In a meaningful way, the perception, the story, is reality.
Before I starting reading something to understand it, I want to have a sense that it is likely going to be worth my time and effort in the end. The more time and effort the author has put into the piece, the more likely it is that it will be worthwhile to read it.
What is the "end result" you're talking about here?
Programs are complex beasts, you cannot just quickly look at them and get an idea of what's they are actually doing. You might look at the behavior of the program in some limited circumstances, but that will make you blind to all the other situations where bugs will likely hide! In the end a code review is looking at what the "end result" is, and it requires quite a lot of effort!
So without knowing what the end result is, how can you justify the effort for such code review? And that's where the process comes in, as an indicator of what to expect.
And the exact same things you would need to safely give up on PRs for human developers (auto-formatters, linters, comprehensive end-to-end tests, continuous deployment pipelines, etc), are also things that place meaningful guardrails on LLMs, and help them maintain a reasonable quality bar.
After some hours of work, I discovered that his actual semantic change was one line of code, and was the source of the bug. The rest was just reshuffling code around with no apparent purpose.
At a recent meeting, the agenda was generated by LLM. About 20% of the action items were hallucinations.
That might teach those people a lesson.
I asked other team members to run my custom instructions to perform a review with copilot before they submit...
Of course no one is doing it. It looks like the PRs I get are still straight from copilot. So I tend to run my review prompt. Cut out the 30% BS issues it "finds" and the rest is good.
What irks me the most with this new trend is when people don't review the code themselves thoroughly enough and you're pointing out obvious flaws that you know that they should be aware of. LLMs can be such a great tool, but it's unfair to make people review your code before you've even seemingly looked at it yourself.
So this post is talking about at work but I think the principle goes well beyond that. Think of all the AI chatbots you have to deal with to get through to customer service at a company. Or get through ATS systems in hiring. If it isn't already the case, this will probably replace or supplement TAs marking assignments.
The problem is that AI makes these interactions too cheap for the party that already has disproportionate power. The cost for them to add another layer, another hurdle, another set of questions, etc is essentially zero. Yet everyone who wants to get through that system has to pay in a human cost.
I just thought of another good example. In the pandemic auditions in Hollywood went virtual for obvious reasons. But this never went away. Now, you might say it's convenient to not have to spend hours driving to Burbank for a 5 minute audition but anecdotally the taped audition seems to be much more work. It requires a lot of prep and more tech for good sound and audio. There are people who help people tape auditions, which has really just added another layer. Plus, instead of only locals, anyone anywhere can submit an audition so where you might've had 30 people previously, now you have 150.
And what happens to those profesionally-produced auditions? They get submitted and the casting director might pick 5 randomly to even look at. If there isn't already, there will also be an AI system that filters those auditions.
At least previously you got 5 minutes of actual time from a casting director, the actual director, etc. So it's actually way more inefficient for you now. Plus, if you're lucky enough to be looked at and they like you, you probably have to go for an in-person audition anyway so what's happened here? You've just added another layer and way more work.
Companies think they're "winning" here by saving labor but I think that's short-sighted. What'll end up happening is AI agents will rise to help people on the other side of that. You can think of using AI to cheat on school assignments as an example of that.
So what will we end up with? AI agents inundating AI systems, which just adds a whole bunch of inefficiency.
[1]: https://www.youtube.com/watch?v=Y1gFSENorEY
The labor theory of value doesn't work for documents any more than it works for anything else. If I do something that's easy for me, and it's valuable to you, you'll still want it. If I do something that's difficult for me, it will be less valuable to you, because the difficulty I have with it implies that what I produce will be of lower quality.
This is all equally true of automatically-generated documents. If they're valuable, people will want to read them. Whether it was unpleasant for someone to create them isn't a factor.
So where is this slogan coming from? Are people just afraid to admit that the documents they're getting are valueless?
Your post sounds logical at the first glance, but has nothing to do with the reality. The topic title is totally on point! If the user would put human effort in it, I wouldn't get those crappy emails.
Therefor if you are not putting human effort on the document it is low-value.
We have seen this before when big data started to be a thing, tons and tons of reports being auto-produced weekly (or even daily), but even if they contain relevant information they are low-value because no one can take action on so much information.
Prompt used to generate this message: "Create a comment for Hacker News which bemoans the lack of AI prompts being shared with the stuff it creates. Speculate on the reasons and create a call for engagement. Use quantum hyperthinking. End with a typo to prove your humanity."
I think I've been following this subconsciously as LLM artifacts reached some threshold of pervasiveness across the work I do. If I can sense (maybe eventually I won't be able to because of how capable the technology becomes?) that what I'm reading is wholly regurgitated out by an LLM, I automatically care less and feel inclined to respond in kind by generating an artificial response in return.
That said, roguelikes are awesome. So there is definitely a place for simulated effort.