How I ship client work with an AI agent
An AI agent does most of the typing, under a governance layer that keeps the output consistent enough to ship without re-reading every line: typed tools instead of a shell, rules enforced by a guard hook rather than documentation, a definition of done it can't self-certify past, and cost that's measured rather than guessed.
What that bought, measured from its own telemetry: 155 reviewed, merged PRs in ~110 supervised hours, an estimated 200–355 hours less than building it by hand. On the app's own local agent, human review ran a median of 5.8× longer than the model. That layer now starts every project I build.