New trend - concern about massive increase in code review load
The Pragmatic Engineer ● Covered by 2 sources
AI now writes tons of code, so engineers are drowning in code reviews instead. Teams are scrambling with new tools and burnout is creeping in.
Based on reporting by The Pragmatic Engineer — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Something shifted back in January. Once Opus 4.5 and GPT 5.4 started churning out more, and better, code across companies, the bottleneck of building software quietly moved. It used to sit in writing code. Now it sits in reviewing it. Gergely Orosz, writing in The Pragmatic Engineer, says director-level engineering leaders have been talking about little else.
The response since February has been a scramble. A wave of dedicated AI code review tools — CodeRabbit, Greptile, Qodo, SonarQube (rebranded as Gitar) — has seen a burst of adoption. Coding harnesses themselves are jumping in too, with Claude Code review, Cursor review and GitHub Copilot review all offering their own take. Even products that never touched code review before, like Sentry's Seer and Linear, are adding it now that they have visibility into the codebase.
Bigger companies aren't waiting on vendors. Uber built Code Inbox, which includes smart assignment features to keep reviews moving and Risk Profiles that flag which changes deserve closer scrutiny. Cloudflare has its AI Code Reviewer, Faire built something called Fairey, and HubSpot has Sidekick. The pattern across these companies is the same: they tried outside tools and found homegrown ones fit their workflows better.
There's a second, harder approach being floated too — verifying code rather than reviewing it line by line. Nice in theory. In practice, nobody agrees on what counts as thorough testing, whether fuzz testing or formal methods belong in the mix, or how to confirm new tests actually exercise the code they're supposed to check. Tying all of that to observability just adds another layer of unresolved questions.
And underneath all the tooling, there's a human cost. Orosz hears the same complaint repeatedly: engineers who used to review carefully now just wave things through when the AI reviewer has nothing to flag, while the ones still reviewing with real intent are getting buried under a pile of AI-generated pull requests. The tools are multiplying. The exhaustion is too. Nobody's cracked the actual fix yet — what exists right now are experiments layered on top of a growing problem.
My take — AI-written commentary, not fact-checked reporting
Throwing more AI reviewers at an AI-generated code problem sounds suspiciously like patching a leak with another leak. The real tell here is the burnout angle — engineers rubber-stamping PRs because the bot said nothing, while the conscientious ones drown in slop. That's not a tooling gap, that's a culture problem, and no amount of Risk Profiles or Smart Assignments fixes people quietly giving up on caring.
Read more about this at: The Pragmatic Engineer