Key Takeaways

  • Quinn Slack eliminated mandatory pull request reviews before merging to main at AMP because human approval queues erase the speed gains of AI agents.
  • Stacking 20 parallel agent tasks creates zero leverage if an engineer spends the afternoon manually reading diffs.
  • High-velocity engineering teams are shifting their primary metric from mean time between failures to a 15-minute mean time to recovery.
  • Automated rollbacks and small, high-trust teams with end-to-end ownership replace bureaucratic pre-merge verification.

The Parallelism Bottleneck

For a decade, engineering managers treated mandatory pull request review as gospel. Nothing touched main without another engineer giving an explicit green checkmark. That rule worked when humans typed out code line by line at human speed. When one developer produces one PR every two days, having a teammate spend 20 minutes reviewing it introduces minimal friction.

AI coding agents broke that math. When an engineer can spin up 20 parallel cloud sandboxes to tackle 20 distinct tasks simultaneously, the bottleneck shifts immediately. If you demand human sign-off on every diff, your team drowns in code reviews.

Slack described this exact tipping point when building AMP: “If we have to review code, then what's the benefit of me stacking up 20 things that are waiting for review? I don't benefit from greater parallelism.”

When agents run long-horizon tasks and verify their own work against test suites, forcing a human to read every line negates the speed advantage. Slack noticed that as agent reliability crossed a baseline threshold, the engineering behavior flipped naturally: “When you see the agent so reliably getting it right enough times and you start to stop reviewing all of the code, then that frees up more of your time to run more things.”

Optimizing for Recovery Over Prevention

Traditional engineering culture obsesses over mean time between failures. Teams introduce staging environments, manual QA passes, and multi-stage approvals to ensure zero bugs hit production. But in fast-moving consumer and end-user software, preventing every single bug creates massive organizational drag.

Slack shifted AMP to prioritize mean time to recovery instead. “Mandatory code review before it gets to main? Yeah, it's dead. It's been dead ever since we started working on AMP,” Slack explained. “It's the mean time before recovery versus the mean time before failure approach, but for end user software that's moving so quickly, that's the right approach for us.”

If you can detect a regression instantly and roll it back within 15 minutes, a minor bug in production costs far less than slowing down your entire development loop. This model requires two structural pillars: automated rollbacks and extreme engineer ownership.

“The most important system is having a team that is really trusted,” Slack noted. “Some of what AI does is it means that you can have a team of people that are more trusted, that have more skin in the game, that have more ownership end to end.”

When developers own their features from prompt to production, they monitor their own releases instead of treating code review as a safety net that absolves them of responsibility.

What to Do With This

Audit your GitHub repository settings this week. Disable required review approvals for non-critical services and internal tools, and replace that rule with automated test suites and one-click rollback scripts. Let your engineers merge agent-generated pull requests directly to main, then measure whether your shipping velocity increases without degrading your incident recovery time.