-
2026-06-27
To make production errors file their own issues, I gave the system an Agent inbox — I'd actually wanted it to fix the bugs
Monitoring dashboards are ugly, limited, and cost money — and this is a side project. In the AI era I wanted to try Agent-first: no dashboard, just an agent watching an inbox in real time. So production errors email a robot inbox and a local daemon opens GitHub issues on its own.
-
2026-06-21
I built a validator to judge whether AI translation has "soul" — then had to validate the validator
The selling point was "translates with soul," but how do you get a machine to judge "soul"? I had an LLM act as judge and score it — and it gave everything a perfect score. The validator has to pass validation first. A lesson on validators and loops.
-
2026-06-13
To understand the multi-agent orchestration Yegge keeps talking about, I hand-wrote a toy orchestrator
Everyone's talking about "multi-agent orchestration." I actually hand-wrote a minimal one and got it running — then found the hard part isn't making them cooperate, it's "knowing they did it right." Open source, you can run and verify it yourself.