My takeaway from Anthropic rebuilding its source control for agent-scale work: the unit of autonomy has to be bigger than a coding task.
Writing a change is one step. A software factory has to carry it through integration, testing and release, then use failures to improve the next attempt.
The hard part is how changes interact. Two can pass separately and fail together. A passing result describes one version, not every version after it. With more agents, keeping that evidence tied to what ships gets harder.
I’d measure how much work reaches production correctly, how often engineers step in to fix or unblock the work, and how quickly the system recovers. Counting agents or generated code doesn’t tell me that.
Faster coding is valuable. The bigger leap is making the whole lifecycle capable of absorbing it.