# code-review

Geoffrey Litt asks a sharp question for the agent era: if agents can write and verify more code by themselves, why should humans still understand the code? His answer is that understanding is not only for verification. It is how humans keep participating.

Code-writing models are 'English to code' interpreters — writing code is no longer the bottleneck; reviewing and merging it safely is. Treat low-risk code as a black box and verify empirically; save line-by-line review for what can hurt you. Claude Code's creator Boris Cherny agrees.

Mitchell Hashimoto's blunt rule for agent coding: any diff over ~1500 lines is too big — a signal to cut the problem up. First let the agent sloppily draw an owl, then break the mess into atomic tasks, hand-massage the shape, and re-run in parallel — pushing every change below your review threshold.

The 2026 data all points one way: AI pushes raw code output up about 4x, but real delivered value only rises about 10%. The gap in between is all review debt. Writing code got cheap; being sure it is right did not. Code review went from a side effect of engineering to its most leveraged front line.

Google Engineering Practices frames code review as code-health work, not a perfection ritual: approve CLs that improve the system, while aligning design, tests, speed, comments, and author habits around maintainability.

Boris Cherny says his team lets Claude Code find 99%+ of bugs first, then an engineer sanity-checks to make sure nothing obvious slipped through.

Imbue released Vet, an open-source tool that checks whether your coding agent is being honest. It reviews conversation logs and code changes, catching agents that claim tests passed when they never ran them. Runs locally, zero telemetry, CI-ready.

Simon Willison's new Agentic Engineering anti-pattern hits hard: do not submit AI-generated code you have not personally verified. That is not saving time; it is stealing reviewer time. The post pairs principles with a terraform destroy horror story.

swyx argues the final boss of Agentic Engineering isn't writing better code — it's eliminating the human Code Review bottleneck. The SDLC is about to flip upside down.

Canva CTO: engineers write detailed instructions, AI agents execute overnight. Senior engineers now 'largely review.' Anthropic CEO calls this 'Centaur Phase.' Few orgs redesigned work for AI. Cora startup achieved 20-30 eng output with 6 people. AI improves exponentially, humans don't.

Drexel/Missouri S&T analyzed 33,596 agent-authored GitHub PRs from 5 coding agents. Overall merge rate: 71%. Codex: 83%, Claude Code: 59%, Copilot: 43%. Rejection cause: no review. LeadDev warns PR flood is crushing monorepos/CI.

Cognition ships Devin Autofix: review bot comments auto-trigger fixes → CI reruns → loop until clean. Humans only step in for architecture calls. Key insight: a single agent is a tool, but agent + reviewer loop is a system — and systems compound.

GitHub's Agent HQ now offers multi-agent support (Claude, Codex, Copilot) for Copilot Pro+ & Enterprise users. Run multiple AIs simultaneously in GitHub/VS Code to tackle problems from different angles. Outputs become Draft PRs. A paradigm shift for code review.