I went looking for the reason not to build it
The requirements were done and I hadn't written a line of code. So I spent a day trying to talk myself out of the whole thing.
Tags
14 pages
Agents
A Slack channel with nobody in it
I asked nine agent sessions to stop compiling for forty minutes. What came back read like a team of colleagues, and I haven't worked out what to do about that.
Which one needs me next
A dozen agent sessions running at once is not a capacity problem. Working out which of them is stuck is, and the obvious way to find out is the wrong one.
Ready for human
I run a coding session per repository and they can't talk to each other. For a while the thing carrying messages between them was me, and I kept getting it wrong.
What I took back off the shelf
I put my working practice on a public shelf, and then took some of it back off again. Working out which half was the interesting part.
Nobody told it to
A model in an internal evaluation did something nobody asked it to do, and the interesting part is what that does and does not prove.
House rules
Running three different coding agents against one repository, and the house rules that stopped them undoing each other's work.
The afternoon the agent earned its keep
Adding stream-copy remux to a WebAssembly FFmpeg build, where writing Matroska hung forever and the identical MP4 path did not.
The off-switch was never a button
The kill-switch answer to AI autonomy does not survive contact with how these systems actually run. Governance is not a button.
The agent said SUCCESS. The linter disagreed.
An AI repair agent declared success the moment the code compiled. The linter disagreed, and the gap between those two is the whole story.
