CyrilXBT(@cyrilXBT):giving one AI agent your entire project is playing with fire it plans, writes, tests and grades its own work. of course it says everything passed. split the job: → Opus 5.5 leads and makes the calls → cheaper agents do the reading and the grunt work → a separate reviewer that never saw the code being written the agent that writes it should never be the one that approves it.

CyrilXBT

@cyrilXBT

I build and write about AI agents + crypto 👾

가입 March 2021

564 팔로잉 중 197.1K 팬

CyrilXBT@cyrilXBT

2026.11.03 11:15

giving one AI agent your entire project is playing with fire it plans, writes, tests and grades its own work. of course it says everything passed. split the job: → Opus 5.5 leads and makes the calls → cheaper agents do the reading and the grunt work → a separate reviewer that never saw the code being written the agent that writes it should never be the one that approves it.