--- title: Code is cheap now. Review is the bottleneck. ‣ harasim.dev description: AI made writing code cheap, but reviewing it stayed slow, this post argues, proposing a system to fix the bottleneck. A Harvard working paper tracking 718... canonical: https://daily.dev/posts/code-is-cheap-now-review-is-the-bottleneck-harasim-dev-scvxknlut twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Code is cheap now. Review is the bottleneck. ‣ harasim.dev | daily.dev og:description: AI made writing code cheap, but reviewing it stayed slow, this post argues, proposing a system to fix the bottleneck. A Harvard working paper tracking 718... og:url: https://daily.dev/posts/code-is-cheap-now-review-is-the-bottleneck-harasim-dev-scvxknlut og:image: https://api.daily.dev/og/posts/scvxkNLut.png og:image:alt: Code is cheap now. Review is the bottleneck. ‣ harasim.dev og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Code is cheap now. Review is the bottleneck. ‣ harasim.dev **[Harasim.dev](https://daily.dev/sources/harasim-dev)** · [@marcinharasim](https://daily.dev/marcinharasim) · 12 min read · 0 upvotes · 0 comments ## Summary AI made writing code cheap, but reviewing it stayed slow, this post argues, proposing a system to fix the bottleneck. A Harvard working paper tracking 718 firms found agents led to 30% more code but PRs took 49% longer to merge. Faros AI data on over 10,000 developers showed 98% more merged PRs but reviews took 91% longer. The author's fix: cap work in progress at a number like 12 PRs, budget PR size to about 500 lines, require an evidence block in every PR, route changes into light, standard and deep review lanes by risk, and use mutation testing (Infection, with a minimum score of 70) to let machines verify what they can. A hard rule stays fixed: no AI ever approves or merges code. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much longer do pull requests take to merge when developers use AI coding agents? Data from Faros AI covering over 10,000 developers shows teams leaning on AI merged 98% more pull requests, but reviews took 91% longer and the typical PR grew 154% larger. A separate Harvard working paper tracking 718 firms found agents led to about 30% more lines of code but 49% longer time to merge, with no clear rise in resolved issues. _Teams weighing AI coding agents against review overhead can track findings like these on daily.dev._ ### What mutation score threshold should I set with Infection for PHP to gate pull requests? A starting point is a minimum mutation score indicator of 70 and a minimum covered mutation score indicator of 80, checked only against changed files using '--git-diff-lines' and '--git-diff-base=origin/main'. These thresholds are presented as a first guess rather than a value proven to suit every project, so they may need tuning per codebase. _Developers setting up mutation testing gates can compare configurations like this on daily.dev._ ### How should I size-limit pull requests in CI to keep AI-generated diffs reviewable? A practical budget is around 500 changed lines per pull request, excluding lockfiles and generated code, enforced with a CI job that fails the pipeline and tells the author to split the PR. The limit should not be raised to accommodate larger AI-generated diffs, since doing so only hides the underlying review bottleneck. _Teams tuning CI gates for AI-generated code can track patterns like this on daily.dev._ ## Similar posts on daily.dev - [AI Code Review Bottleneck 2026: Why PRs Are Stuck in Review](https://daily.dev/posts/ai-code-review-bottleneck-2026-why-prs-are-stuck-in-review-o7qcycgq0) · Alex CloudStar · 2 upvotes · 1 comments - [Why AI Code Reviews Are Slowing Down PRs](https://daily.dev/posts/why-ai-code-reviews-are-slowing-down-prs-yubjqxjqr) · Aviator Blog · 0 upvotes · 1 comments - [Agentic Code Review](https://daily.dev/posts/agentic-code-review-0ue8fcsa1) · Addy Osmani · 147 upvotes · 14 comments --- Tags: [#cicd](https://daily.dev/tags/cicd), [#ai-coding](https://daily.dev/tags/ai-coding), [#code-review](https://daily.dev/tags/code-review), [#github-actions](https://daily.dev/tags/github-actions), [#mutation-testing](https://daily.dev/tags/mutation-testing) [View this post on daily.dev](https://daily.dev/posts/code-is-cheap-now-review-is-the-bottleneck-harasim-dev-scvxknlut) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Code is cheap now. Review is the bottleneck. ‣ harasim.dev","url":"https://daily.dev/posts/code-is-cheap-now-review-is-the-bottleneck-harasim-dev-scvxknlut","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/code-is-cheap-now-review-is-the-bottleneck-harasim-dev-scvxknlut"},"datePublished":"2026-11-03T09:36:57.591Z","dateModified":"2026-11-03T11:21:54.987Z","description":"AI made writing code cheap, but reviewing it stayed slow, this post argues, proposing a system to fix the bottleneck. A Harvard working paper tracking 718...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b62d9c086464ac022a1dbb2f3c768a15?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/b62d9c086464ac022a1dbb2f3c768a15?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Harasim.dev","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Person","name":"Marcin Harasim","url":"https://daily.dev/marcinharasim","image":"https://media.daily.dev/image/upload/s--KtGCHUBv--/f_auto/v1768764027/avatars/avatar_oLOnIowq6Jt0VKlbUjH6v?_a=BAMAMiiu0","description":"Building resilient PHP systems and high-performing teams through architecture and mentorship.","interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"EndorseAction"},"userInteractionCount":940}},"commentCount":0,"discussionUrl":"https://daily.dev/posts/code-is-cheap-now-review-is-the-bottleneck-harasim-dev-scvxknlut","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"cicd,ai-coding,code-review,github-actions,mutation-testing","timeRequired":"PT12M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Harasim.dev","item":"https://daily.dev/sources/harasim-dev"},{"@type":"ListItem","position":3,"name":"Code is cheap now. Review is the bottleneck. ‣ harasim.dev"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/code-is-cheap-now-review-is-the-bottleneck-harasim-dev-scvxknlut#faq","mainEntity":[{"@type":"Question","name":"How much longer do pull requests take to merge when developers use AI coding agents?","acceptedAnswer":{"@type":"Answer","text":"Data from Faros AI covering over 10,000 developers shows teams leaning on AI merged 98% more pull requests, but reviews took 91% longer and the typical PR grew 154% larger. A separate Harvard working paper tracking 718 firms found agents led to about 30% more lines of code but 49% longer time to merge, with no clear rise in resolved issues. Teams weighing AI coding agents against review overhead can track findings like these on daily.dev."}},{"@type":"Question","name":"What mutation score threshold should I set with Infection for PHP to gate pull requests?","acceptedAnswer":{"@type":"Answer","text":"A starting point is a minimum mutation score indicator of 70 and a minimum covered mutation score indicator of 80, checked only against changed files using '--git-diff-lines' and '--git-diff-base=origin/main'. These thresholds are presented as a first guess rather than a value proven to suit every project, so they may need tuning per codebase. Developers setting up mutation testing gates can compare configurations like this on daily.dev."}},{"@type":"Question","name":"How should I size-limit pull requests in CI to keep AI-generated diffs reviewable?","acceptedAnswer":{"@type":"Answer","text":"A practical budget is around 500 changed lines per pull request, excluding lockfiles and generated code, enforced with a CI job that fails the pipeline and tells the author to split the PR. The limit should not be raised to accommodate larger AI-generated diffs, since doing so only hides the underlying review bottleneck. Teams tuning CI gates for AI-generated code can track patterns like this on daily.dev."}}]} ```