Your team’s code review process is probably an expensive illusion of quality. Here’s what the evidence says, and what to do instead.

TL;DR

  • Pull requests were designed for open source contributions from untrusted strangers. Applying them to trusted teams is a category error.
  • Peer-reviewed research shows code review’s primary value is knowledge transfer, not bug detection. Less than 15% of review comments relate to actual bugs.
  • Async PR workflows mean your code spends 86-99% of its lead time waiting. One organisation spent 130,000 hours in a single year waiting on PRs that received zero comments.
  • DORA research across 36,000+ professionals shows trunk-based development correlates with dramatically higher software delivery performance, and faster code reviews alone improve performance by 50%.
  • The alternative is T*D: Test-Driven Development (build quality in), Trunk-Based Development (integrate continuously), and Team-focused Development (review during creation, not after).
  • The transition is gradual: optimise PRs first, adopt Ship/Show/Ask, then move to pairing and trunk-based development as trust and automation mature.-
  • Tamo240@programming.dev
    2·
    19 days ago

    Like, you might have someone working on something in a new codebase but misunderstanding the architecture or just going in the wrong direction in general and a review is the best way to correct course before getting too far in

    I think the article is suggesting this person should be pair programmed with until they understand the architecture and can be trusted to contribute correctly, and I actually kind of agree - it always feels terrible to tell someone a PR they’ve worked on possibly for days is completely the wrong direction, and arguably this is already ‘too far in’ if they’re going to need to essentially start again.

    Intervening earlier in the process should lead to less wasted effort overall, but people often seem to treat pair programming like its two people at 50% efficiency, when it actually saves a lot of cycle time on reviewing code.