Catch-up — Review by Risk, Not by Whether a Human Read Every Line
Code review should match the risk of a change, not satisfy a blanket rule that a person reads every line.
- Anthropic says merged pull requests to the Claude Code codebase more than doubled from June to September while the reverted share fell by over 40 percent.
- Anthropic argues that asking whether a human read code is an imprecise proxy for matching possible failures to automated or human checks.
- Anthropic says the share of internal pull requests receiving substantive review comments rose from 16% to 54%.
- Anthropic says an agent's approval never counts as a person's approval in its review process.
Catch-up written 24 October 2026 for 2026-10-16; covers a source published 2026-10-15.
A uniform review rule spends scarce attention on low-risk changes and can still miss the specific checks that dangerous changes need.
Choose three risk tiers for repository changes. Assign automated checks and human approvals to each, then monitor reversions to see whether the policy is too loose.
Put it to work 30 minutes
A three-tier, risk-based code-review policy
Draft a three-tier review policy: example paths for low, medium, and high risk, required human approvals, and the revert-rate signal that makes you tighten it.
See a worked example
Use a real, bounded workflow and record unknowns rather than estimating them. Keep the result small enough to review with a teammate.
- Workflow or decision
- A real workflow the team repeats each week.
- Evidence to record
- Quality, time, cost, failures, and reviewer notes.
- Decision rule
- Keep the current approach unless the agreed checks pass.
Your turn
Name the exact case you will test.
List the observations needed to make the decision.
Write what the evidence will make you do.
Your draft stays in this browser. No account needed.
These are Anthropic's figures for its own repository and tools. Teams without merge queues, flags, and on-call coverage may not be able to reduce human review safely.
Reported facts and quotations were checked against the named sources; the recommended exercise is editorial guidance to test in your own context.
Did the risk tiers reduce review work without increasing reversions?
Watch: Revert rate by risk tier · Substantive review findings and escaped incidentsKeep Going