RE: https://tldr.nettime.org/@tante/117172155532057446
"agents.md" files "work" because they look so meaningful to humans which means adding them to the repo looks as if you explained something relevant. The misunderstanding is that "the agents reads the file and understands it". The file is just slapped into the prompt to manipulate which tokens it generates. That's why you can't even guarantee that the "tool calls" in agents.md files are actually being made: It's a stochastic process, you might be unlucky and the dice fell in a way that lead to something else happening.
Researchers at ETH Zurich show that "Agents.md" files don't do much. "AI" is cosplay all the way down.
(Original title: Your AGENTS.md file doesn’t actually do anything)
https://pivot-to-ai.com/2026/08/27/your-agents-md-file-doesnt-actually-do-anything/
