The useful question is not whether the agent wrote code. It is what evidence makes the code safe to ship.
Seeing an event and successfully applying an event are two different states.
A census of the failure surface, taken from a capability that has been built rather than imagined.
A progress report with the real counter, four bugs that only appear under concurrency, and the parts that are not built yet.
The happy path is increasingly cheap. Production readiness lives in everything around it.
