Source: Engineering blog post 'How the Claude Code team does code review' by Addy Osmani, Morgan Westlee Lunt, and Qing Wang, published on claude.dev on October 15, 2026.
As agentic software generation tools proliferate, developer teams face a fundamental bottleneck: code production rates outstrip the capacity of human engineers to conduct line-by-line reviews. At Anthropic, daily merged changes across the central monorepo surged from approximately 2,500 in May to more than 30,000 in September, creating severe coordination pressures in the standalone Claude Code repositories.
To address the volume, the Claude Code team transformed review from a sequential end-of-branch human inspection gate into continuous automated filtering. Before a pull request is opened, an authoring agent must pass local test, build, and lint suites, followed by a second automated agent evaluating the generated diff in an isolated, fresh execution context.
Once submitted, every pull request triggers required multi-platform CI alongside two distinct agentic inspection pipelines: a general code review pass where specialized agents hunt for functional defects in parallel and submit findings to a secondary panel of verifier models that vote on issue validity, and an independent security review pass. Pull requests are subsequently classified into risk tiers: low-risk changes require zero human approvals, while sensitive architectural paths mandate one or two human sign-offs.
Merged code enters an automated merge queue where validation runs against the latest main branch before deploying behind internal feature flags. The authors emphasize that while automated layers dramatically reduced incident rates, human engineers remain essential for high-level system architecture, business trade-offs, and unexpected cross-system failures.