A weird experiment I've been trying the last few weeks is having Claude take ove

Boris Cherny describes an experiment in which recurring Claude routines proposed maintenance changes to the team's apps. The first-person LinkedIn account reports 388 pull requests opened and 180 merged after Claude Code Review and human review.

One routine illustrates the feedback process: it removes code known to be unreachable, but adds logging to code merely suspected of being unused before deciding whether to remove it later. When routines produce poor changes, the team asks Claude to tune them for subsequent runs. The reported workflow combines repeated maintenance work, evidence gathering and review before acceptance.