Killing code review is only safe for small, agent-native teams

AMP, Quinn Slack's 20-person startup, has removed mandatory code review before merge. The bet: a small, trusted team with end-to-end ownership plus AI agents produces better software than a review gate. This is an explicit MTTR-over-MTTF stance — acceptable for fast-moving end-user software, not universal.

The bigger shift is orbs: agents running remotely in cloud sandboxes, letting developers run many tasks in parallel and shut their laptops. Slack argues local dev is dead and that scoped, time-limited cloud access is more secure than agent-on-laptop with broad credentials. That holds only if teams build proper OIDC boundaries.

He also questions CI: if the agent runs tests in a reproducible sandbox, why wait on a deterministic pipeline? Sandbox infrastructure is booming, but this conflates test execution with project-wide regression safety. Agents cannot see what is outside their context; CI catches exactly that failure mode.

Underneath is an organizational claim: a profitable 20-person team can skip process and use agents for marketing, ops, even GitHub. That is a credible small-company strategy, but the 'code review is dead', 'CI is dead', 'local dev is dead' framing is provocative extrapolation from one company.