Coding used to be the bottleneck and isn’t anymore so now there is a lot of brain power being spent at companies to play bottleneck whackamole while tokens are cheap. Problem: code review Multiple levels of automatic code review of AI written code doesn’t really increase trust, and AI code review still lacks the context behind WHY. So there’s all this effort to document everything by adding MCPs to every context store: jira, confluence, gmail, slack, zoom calls, google docs, whatever. The goal is to get to the first approve/reject recommendation before a human even looks at it. Problem: human time Human time is precious, your teams are overloaded, everyone is busy, and has their own brain split between 15 things, so let's intelligently @ people directly in public channels. Round robin it between reviewers, add leaderboards, daily reminders, and some public shaming. But wait. Someone says- can we just…skip human review? Problem: defect escape Whether or not we got a review from a human (rubberstamp or meticulous), we still need to prevent defect escape. Auto merge approved code to main which deploys to staging. Force it to bake on staging for X hours. Run every test suite against it (nothing new). Have adversarial agents use headless browsers to try and break your stuff. Should probably fix things too? Problem: blast radius This one isn’t new either. If you’re going to yeet out code that no human has looked at, you should probably reduce your blast radius with progressive deployments. Squeeze out a little bit, have an agent watch your error logs and dashboards, set a confidence threshold for rollback vs perseverance. Problem: incident triage If something breaks at 3am, unconnected to an ongoing deployment, you shouldn’t have to wait for a human. Have an agent triage the issue and link to relevant dashboards before the human can roll out of bed to get to their laptop. Problem: conflicts Your PR queue is a traffic jam. Agents will need to be constantly rebasing, resolving conflicts, and regenerating changes as main moves underneath them. Problem: trust boundaries How much rope can we give our agents? Under what conditions can we let yhem act without asking permission? Whose agent is going to be the first to screw up and leak PII? Gotta wrap it up, my toddler is waking up. What did I miss?