Harness: agent-ready SCM + AI review gates when PRs outrun humans

Juveria Kanodia and Colin Chartier (Harness; 27 Aug 2026) name the bottleneck that shows up once agents, not humans, set the commit rhythm: legacy SCMs assume a person opens a PR and a few colleagues review it over hours or days. Agents open dozens of PRs in an hour, each touching hundreds of files, at any hour. Indexing lags; diffs pile faster than anyone can read; permissions built for fixed human teammates have no clean answer for a non-human identity that might merge on its own. Fix the repository without fixing review and you get a faster generator nobody can verify in time; fix review without fixing the repository and you review faster on a store that can’t keep up. Both have to move.

Their pairing: Agent-Ready Harness Code Repository (SCM claimed scale-tested for thousands of PRs/commits per second, RBAC/OPA-scoped agent identities that inherit from the triggering developer) and AI Code Review inside the same Software Delivery Agent sequence. Review looks at risk-grouped diffs rather than file-by-file scrolls, grounded in introduced-change risk. Customizable AI Checks sit as merge gates at account/org/project with inheritance; required failing checks cannot be squash-merged. Build, test, and secure cannot depend on a human kicking them off once agents generate most PRs. The portable claim is governance moved earlier: scoped access before merge, verification gates that agents cannot waive by volume alone.

Why it matters for code review

Always-on factories only compound past peak-session heroics when the review surface scales with overnight volume — identity scope, risk-grouped diffs, and non-waivable checks — not when humans pretend to keep reading every line. Continuous harnessed momentum without that surface is just FOMO with a larger backlog. Bursty Claude mastery still wins when a single change needs deep judgment the gate cannot encode; agent-scale SCM is for the fleet that would otherwise drown the morning review queue.