AI Burnout Is Real and Getting Documented
A HN thread on AI-assisted engineers burning out surfaced a specific, uncomfortable diagnosis: writing code was slow but you understood what you built, while reviewing AI code is fast but you accumulate blind spots. The thread on people throwing AI-generated walls of text into conversations made the same point from a different angle. People are sending screenshots of ChatGPT conversations to colleagues as if that constitutes analysis. People are receiving 7-page AI-generated proposals from their chief data officers and having to figure out what to do with them.
The 'Tell HN: I'm tired of AI-generated answers' thread went long and got specific. A recurring pattern in the comments: the signal-to-noise problem has inverted. It used to be hard to find information. Now it is hard to find a human who has actually thought about something. Several commenters noted they have started explicitly stating 'do not send me AI-generated analysis' in professional settings.
The nuance here is that the people complaining loudest are not anti-AI. They are anti-slop. The distinction matters. Nobody in these threads is arguing for going back to doing things slowly. They are arguing for maintaining the expectation that someone, somewhere, actually thought about the thing before sending it. That expectation is eroding fast.
So what?
For founders managing teams, this is a real productivity trap disguised as a productivity gain. Engineers who review more AI code than they write are losing depth, and that debt will show up in your codebase at the worst possible time. Consider establishing explicit norms about when AI output needs to be owned vs. forwarded.