A Faros AI study of 22,000 developers across 4,000 teams — published in March 2026 and surfaced widely through Addy Osmani's "Agentic Code Review" on O'Reilly Radar — put a precise number on something engineering teams have been feeling for months: median PR review duration is up 441.5% since coding agents became mainstream. PRs merged with zero review are up 31.3%. Nobody chose to stop reviewing. The volume just outran the capacity.
The structural reason is straightforward. Code review used to be self-regulating — a senior engineer could read code faster than a junior could write it. Agents removed that speed advantage. A diff arrives instantly; understanding whether to trust it does not. Compounding the problem: agents tend to discard their reasoning at diff time, leaving reviewers to reconstruct intent from the code itself, which is measurably harder and slower.
A companion 2026 paper studying 33,707 agent-authored PRs found that agents tend to abandon PRs when they receive subjective feedback — reviewer abandonment accounted for 38% of rejected agent PRs. The implication is that review queues are filling with PRs that will require significant back-and-forth, and that time cost falls entirely on human engineers.
For teams working across multiple repositories, this dynamic is compounded by a visibility problem that gets relatively little attention. A single feature in a microservice architecture might touch a payments service, an auth layer, and an API gateway — three separate repos, potentially on two different providers. Each PR looks reasonable in isolation. Together, they represent a coordinated change that can only be assessed properly by seeing all three at once. When the review surface is three separate browser tabs on three separate dashboards, the cognitive overhead alone is enough to push reviewers toward skimming or skipping.
The practical interventions that the research points to are: risk-scoring PRs before a human opens them (patch size and file type are cheap, early signals), fast-tracking low-risk agent PRs, and raising the bar for what gets full review. All of that becomes significantly easier when the full picture of in-flight PRs is visible in one place rather than scattered across repositories.
For multi-repo teams trying to manage this, Code Board aggregates PRs from GitHub and GitLab into a single board with AI-powered risk scoring and cross-repo context — making it possible to triage the agentic PR flood by priority rather than by arrival order.
Top comments (0)