Your agents keep making the same mistakes. Nobody has time to fix it. AI coding agents are getting better at the tasks you give them direct feedback on. Everything else stays broken. You leave the same comment in code review three sprints in a row. There's a recurring task that could run as an automation but it's on the backlog because no one has time to stop and systematize it. The context your agents need to do better work — updated conventions, patterns from past PRs, recurring fixes — ...