Who reviews the code now?

Agents made writing cheap. Review was supposed to be where humans stayed in the loop. This week five people share about what is happening instead, and what they’re doing about it.

• LEADERSHIP · Anton Zaides
  The broken windows theory of coding agents (opens in a new tab) · the confession: two reviewers per PR to zero, in five months
• DISSENT · Remus Lazar
  Two agents, not five (opens in a new tab) · review didn’t die, it moved to Slack and left no trace
• PROCESS · Devrim Ozcay
  My staff engineer quietly stopped reviewing AI-generated code (opens in a new tab) · three lanes by blast radius; submit an invariant, not a description
• ON CALL · Sylvain Kalache
  AI handles incidents, engineers lose touch with their systems (opens in a new tab) · the same collapse, one layer down: the skill that atrophies is the 3am one
• THE MATH · Brian Curry [paid]
  Building an autonomous coding fleet that actually scales (opens in a new tab) · review everything and you sustain one agent; gate 15% and you sustain thirteen