Background agents
The key bottleneck has become clearly communicating well-defined specs that long-running background agents can actually iterate on. It’s possible the constraint is on the execution side, but it feels like my communication and ability to spec is the real limiter. Worth messing around with Pi to see if a custom harness is more conducive.
Padmi — worktree bug bash
Thinking about using worktrees for the test harness — N games running in parallel, each finding bugs independently, with an orchestrator that merges. A full bug bash. That would be sick.
praxis
Still concerned about quality, especially the memos. This is clearly agent drift — it picks up my thoughts but needs a human touch. Excited to see progress on the forum and reliability improvements though, that’s nice.
Day
Not as productive as I’d like tbh. Went to get a haircut, exhausted from the morning run. Try and reset, do better tomorrow.