Harnesses, evals, review gates, tool design, and the failure modes that show up next to them — notes for teams shipping real software with coding agents. Updated as the practice moves.
A four-phase goal loop that eliminates most rework from agent-driven development with Codex.
The useful layer around coding agents is shifting from prompts to repeatable environments, gates, and workflows.
A useful thread on making command-line tools legible to coding agents, not just convenient for humans.