People who previously wouldn't touch a README.md with a ten foot pole are now writing entire novels in Markdown for their AI tooling.

@frederic Yes. People who couldn't be arsed to write a single word of documentation for their human colleagues suddenly don't have any issue at all with writing miles and miles of instructions for their AI colleagues.

Same for people who previously couldn't find time to mentor a student or new junior colleague because "it's too much work to describe the task in enough detail". For their LLM? No problem at all. 🤷
That's the one that really gets under my skin.

@mmeier @frederic bugs the shit out of me too

@mmeier @frederic That's curious, because that aspect doesn't seem to bug me.

I'm the person who writes documentation of techniques I've derived and shares them with colleagues - none of whom ever seem to return the favour.

However, they will happily video chat about or screen-share what they've done, so it's not the sharing part that puts them off.

My guess is that commitment into stored text scares them, plus the risk of it being critiqued - and they know the LLM won't do that.

@mmeier @frederic

They're showing you who they are. They don't care about other people and the LLM is serving as unpaid slave labour that does their bidding without question.

This is a recurring theme with AI and it's going to get worse.

@mmeier @frederic

Well … at least this documentation is also valuable for humans. And LLMs will have plenty of time to explain it to everyone, according to the asker’s level.

@mmeier @frederic I keep seeing this, and I can only imagine that like, half of it is generated anyway, and the rest is the "one more hit bro" effect of the dopamine feedback loop that generating output so rapidly causes, so they just keep adding more and more and more and then they're used to doing whatever to generate that feedback loop.

Couple with how real people don't give the same "you're absolutely right!"-style feedback and repeated exposure to doing this and it becomes their default. Unfortunately.

Welp, they keep vibe coding their shit and they'll keep ending up with music hosting servers that have a paid component but isn't shipped as a separate binary and requires a single line or 2 of code to be patched to enable for free.

@mmeier @frederic and then these people don't even actually write the documentation. they go "claude, document it" and the documentation is slop
@mmeier @frederic To be honest, in my anecdotal experience, these instructions are, in turn, typically LLM-generated themselves. Which is a whole other bag of worms.
@rabbit @frederic Oh yeah, that's another whole bag of weird I might just be too "left behind" to fully grasp.
@rabbit @mmeier @frederic With (uninformed) vibe coding it’s worms all the way down.
@mmeier @frederic Yes! I accidentally read one of those agent-files in a project, and was ALMOST impressed by how comprehensive it all was (at least there) in terms of architecture, style, and even comment tone, because that's obviously useful. And then I realized that they'd probably scream if an actual prospective contributor read it for advice...

I guess I am guilty of not writing sufficient documentation for some of the code I have written. But I have definitely taken the time to explain to new colleagues how it works. And I have also done many pair programming sessions with colleagues to improve the code. And they have now written documentation of the code for AI to read.

But the documentation they wrote actually looks good, so that can probably be helpful for new hires as well. We haven’t stopped hiring, we have hired a couple of new people this year.

@mmeier @frederic

Org incentives. Teaching others or writing documentation was seen as unnecessary and people that were good at it were usually let go for failing to meet KPIs.

Now with the AI hype a high token use - aka. wasting time on writing even the most stupidest stuff to the AI - is the KPI...

@mmeier @frederic I can't stop thinking about this because it's so true, I just watched my coworker navigate to a new repo and immediately open AGENTS.md before anything else because no one maintains the README.md 😭