Omer Bin SaleemBuild with AI Subscribe
← All research AI · Code4 sources

AI made code faster. Review is now the bottleneck.

Median time in code review went up 441% between each organization’s lowest and highest AI-adoption periods.

Slide 1 of 9: AI made code faster. Review is now the bottleneck.Slide 2 of 9: AI made code faster. Review is now the bottleneck.Slide 3 of 9: AI made code faster. Review is now the bottleneck.Slide 4 of 9: AI made code faster. Review is now the bottleneck.Slide 5 of 9: AI made code faster. Review is now the bottleneck.Slide 6 of 9: AI made code faster. Review is now the bottleneck.Slide 7 of 9: AI made code faster. Review is now the bottleneck.Slide 8 of 9: AI made code faster. Review is now the bottleneck.Slide 9 of 9: AI made code faster. Review is now the bottleneck.

The numbers

  1. Median time in code review+441%

    comparing each organization’s highest AI-adoption periods with its lowest.

    Measured from tools, not a survey. Vendor data, and a correlation.

    Source: Faros AI, AI Engineering Report 2026, 22,000 developers

  2. Same report

    Output went up. So did the bugs.

    Epics per developer+66%more work completed
    Bugs per developer+54%lowest vs highest AI-adoption periods

    Source: Faros AI, AI Engineering Report 2026

  3. Same report+31%

    more pull requests merged with no review at all.

    Source: Faros AI, AI Engineering Report 2026

  4. 8.1 million pull requests

    AI PRs sit in the queue longer

    Wait before review4.6xlonger than a manual PR
    Review once picked up2xfaster than a manual PR

    Once opened they move faster. Getting opened is the slow part.

    Source: LinearB, 2026 Software Engineering Benchmarks (vendor data)

  5. A Google engineer, in Google’s own DORA research
    “Reviewing [another’s] code is so much harder than writing it.”From 1,110 open-ended answers by Google engineers

    Source: DORA, Balancing AI tensions, 10 Mar 2026

Fix the review, not the writing

  1. Let AI take the first review pass. I run an AI reviewer on every merge request, about 600 so far. People then review the design. GitHub says AI code review use grew 90%+ in a year.
  2. Track review time next to throughput. PR count shows what shipped. Wait time shows what it cost.
  3. Keep PRs small. An agent can write a thousand lines in a minute. Nobody reviews that well.

Sources

  1. Faros AI, AI Engineering Report 2026 (vendor telemetry, 22,000 developers), 12 Apr 2026
  2. LinearB, 2026 Software Engineering Benchmarks (vendor, 8.1M PRs), 2026
  3. DORA, Balancing AI tensions, 10 Mar 2026
  4. GitHub Octoverse 2026, 28 Oct 2026

Is review your bottleneck yet? Tell me on LinkedIn or Instagram.