Bain says code review is absorbing the AI coding speedup

Bain's 2026 Technology Report finds developers finish about 21% more tasks with AI while review time rises about 91%, making human review the constraint.

Bain's Technology Report 2026 says AI coding speedups are being absorbed downstream by code review. In its chapter on agentic software development, published September 29, 2026, Bain reports that developers complete roughly 21% more tasks while review time rises about 91%. Developers are also juggling 47% more concurrent workstreams. If you ship AI-built code, the slow step is now verifying it, not writing it.

The figures come from Bain's Tech and Engineering Survey of 293 senior technology leaders. They reflect what leaders report and expect, not measured developer telemetry. Respondents expect a 148% improvement in release-cycle speed and a 95% productivity uplift within one to two years. Today they capture gains of 20–27% across key metrics, numbers also reported by Telecom Review Americas. Bain does not say how much of that gap is caused by review alone.

Bain's recommendation is to "automate confidence." That means a deterministic harness with risk scoring, policy-as-code guardrails, evaluation frameworks and audit trails, so humans focus on exceptions and higher-risk changes. Bain cites Amazon, where AI handles up to 70% of code reviews. It also cites Stripe, which merges 1,300 AI-authored pull requests a week, all going through normal human review.

Others describe the same bottleneck. QA.tech, a testing vendor, cites Sonar's January 2026 survey, in which 38% of developers said reviewing AI code is more work than reviewing a colleague's. QA.tech argues that verification belongs in the running product, on the pull request. CodeAnt AI, another vendor, recommends smaller PRs and tracking review lag alongside DORA metrics. That approach is cheap, but it doesn't add review capacity.

Still unclear: Bain offers no data on which tactics actually recovered the lost gains.

Sources