Ask HN: How do you keep agent-written PRs reviewable?
Hi HN, I am curious what your teams' best practices are for facilitating human-agent collaboration on software teams.
I want to take advantage of coding agents, but since my team works on robotic software, everything still needs to be human-reviewed. Some folks submit very voluminous and needlessly verbose PRs that are challenging to review. 1. Ruthlessly keep PRs small (use stacked PRs if needed) 2. Very strict CI (tests, linting, if you can afford it an agent to review the code) 3. Related to the first point, but simply refuse to review PRs over a certain number of lines of code (within reason, some PRs cannot be split into smaller ones)