Every pull request I open gets an automated code review from an AI agent before I merge it. It reads the diff, checks it against my project's conventions, and posts its findings as a comment, like a second set of eyes that never sleeps. I run a whole studio's worth of repos solo, so that reviewer is load-bearing infrastructure. I call it the Grid Review Runner.
One weekend it went quiet, and I didn't notice for two days.
A pull request sat in one of my repos with no review on it. I was on the road. When I finally opened my laptop, I had no idea whether the reviewer had looked at the PR and decided it was fine, or whether it had never been told the PR existed in the first place. There was no log to check. There was no signal either way. The review just... wasn't there.
That ambiguity is the whole story. The way the system worked back then, GitHub was supposed to push a notification to my reviewer the instant a PR opened (a webhook, in the usual web plumbing sense). If that push went missing, the reviewer never woke up, and nothing told me it had been skipped. Did the reviewer decline, or did the notification never fire? I couldn't tell. The runner worked beautifully when I was sitting at my desk and fell apart the moment I wasn't. So I rebuilt the transport from the ground up. This post is about that rebuild, and the one architectural decision underneath it: stop waiting to be told about work, and go looking for it instead.
