Search for a command to run...
Series
Notes from daily work with AI agents: how they fail, how I check them, and what a second model or a judge does and doesn't catch. Written from real incidents, not from the demo.
I use Claude and GPT to review each other's work. Adding Jev did not replace that arrangement. It gave it another instrument. The useful distinction is between an agent reviewing a whole change and a
My previous article ended with the human side of an AI failure mode. The system keeps working. Its outputs look good. Checking starts to feel like extra work with no visible return, so one check is sk
A lawyer told me his job was safe because his work never becomes training data. Mine is all data — so I went looking for what's actually left.
For two days I worked with a coding agent that told me it was one model. My editor said it was another. When I finally pushed it to check, it read its own session log and returned this: 263 "model":"
I called GPT a nerd on Reddit. Someone replied that GPT is a rogue. We were both right — and understanding why changed how I think about choosing AI models.
