Search for a command to run...
Series
Real-world notes from building and evolving software development agents: what I tried, what failed, what worked, and the engineering decisions that followed. Focused on production workflows, reliability, verification, guardrails, review, and autonomous ticket-to-PR development.
General-purpose coding agents demo well and fail quietly on a real backlog. So I built a narrow one: one product, one pipeline, no claim without proof. Here is what it does, how, and why. What this s
Repro video, root cause, fix, verification. Every artifact a reviewer would ask for — and every one of them about a branch that was never going to ship.
A dead recording server, a silent stand-in, and a setting that had been lying to me since day one.
Several agent sessions on one machine, each with its own checkout, all sharing one running application. One session's startup replaced another's, and my verification measured the wrong build three times.
