loosing track of whats actually deployed since AI writing most of the code
When I wrote the code, I had an intuitive sense of what was deployed where. scanning the diff, and saying "yeah, this touches the auth flow, keep an eye on login metrics," and feel reasonably confident about what we were pushing (3~4 releases per week).
Now with few agents doing a significant chunk of the coding, the PRs are bigger, the diffs are longer, and I'm approving things based on their design, not so much their actual code.
Everything is “green” until it breaks - tests pass, the logic is sound, features released but now i find myself more and more “reconstructing” the “what actually changed in that release”.
I know that I was supposed to feel more in control, but now I'm less certain about everything that I actually coded.
Tried to improve PR messages - nobody reads them (me included)
Tried to tag with the git SHA - now I have a clear starting point to the endless diffs and messages
Anyone else experiencing this and how are you handling the traceability side specifically? Not asking about code review (agents?!?), more about how you document what actually shipped.