--- title: Study Finds AI-generated Code Fails to Boost... description: A Harvard study of 718 firms finds AI coding tools boost code volume but not completed work. Based on 300 million work events from 700,000 workers between... canonical: https://daily.dev/posts/study-finds-ai-generated-code-fails-to-boost-productivity-due-to-downstream-bottlenecks-zk3tkxmhn twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Study Finds AI-generated Code Fails to Boost Productivity Due to Downstream Bottlenecks | daily.dev og:description: A Harvard study of 718 firms finds AI coding tools boost code volume but not completed work. Based on 300 million work events from 700,000 workers between... og:url: https://daily.dev/posts/study-finds-ai-generated-code-fails-to-boost-productivity-due-to-downstream-bottlenecks-zk3tkxmhn og:image: https://api.daily.dev/og/posts/Zk3TKxmHn.png og:image:alt: Study Finds AI-generated Code Fails to Boost Productivity Due to Downstream Bottlenecks 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. # Study Finds AI-generated Code Fails to Boost Productivity Due to Downstream Bottlenecks **[InfoQ](https://daily.dev/sources/infoq)** · 3 min read · 2 upvotes · 0 comments ## Summary A Harvard study of 718 firms finds AI coding tools boost code volume but not completed work. Based on 300 million work events from 700,000 workers between January 2021 and March 2026, it shows AI agents raised lines of code by 30% but resolved issues by only 12%. AI assistants barely moved the needle, with only the effect on commits being statistically significant. The gap traces to code review: PR review time rose 49%, comments per PR rose 53%, and the share of PRs needing changes doubled, even though AI review tools didn't help. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Does using AI coding agents actually increase completed software output, not just code volume? Not proportionally. A Harvard study of 718 firms using Jellyfish data found AI agents increased lines of code by 30%, commits by 20%, and pull requests by 23%, but resolved Jira issues rose only 12%. AI coding assistants like Copilot showed smaller, mostly non-significant gains and did not increase resolved issues at all. _Anyone weighing AI coding tools against real throughput gains can track findings like this on daily.dev._ ### Why doesn't more AI-generated code translate into faster software delivery? Because code review becomes a bottleneck. The Harvard/Jellyfish study found that as AI increased code output, average PR review time rose 49%, comments per PR rose 53%, and the share of PRs requiring changes doubled, while the share of workers doing reviews grew 14% with no help from AI review tools. _Teams rethinking review workflows around AI-generated code can follow this research on daily.dev._ ### What data did the Harvard AI productivity study analyze? It analyzed about 300 million work events from 700,000 workers across more than 700 firms, spanning January 2021 to March 2026. Data came from HR systems, code version control, issue trackers, scheduling tools like Google Calendar, and generative-AI tools including GitHub Copilot, Cursor, and Claude Code, using the Jellyfish analytics platform. _Developers curious about the evidence behind AI productivity claims can dig into studies like this via daily.dev._ ## Similar posts on daily.dev - [AI productivity gains are 10%, not 10x](https://daily.dev/posts/ai-productivity-gains-are-10-not-10x-mczno6zn8) · Engineering Enablement · 86 upvotes · 30 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#github](https://daily.dev/tags/github), [#productivity](https://daily.dev/tags/productivity), [#claude-code](https://daily.dev/tags/claude-code), [#code-review](https://daily.dev/tags/code-review) [View this post on daily.dev](https://daily.dev/posts/study-finds-ai-generated-code-fails-to-boost-productivity-due-to-downstream-bottlenecks-zk3tkxmhn) ```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":"Study Finds AI-generated Code Fails to Boost Productivity Due to Downstream Bottlenecks","url":"https://daily.dev/posts/study-finds-ai-generated-code-fails-to-boost-productivity-due-to-downstream-bottlenecks-zk3tkxmhn","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/study-finds-ai-generated-code-fails-to-boost-productivity-due-to-downstream-bottlenecks-zk3tkxmhn"},"datePublished":"2026-10-22T14:04:34.890Z","dateModified":"2026-10-22T14:14:51.084Z","description":"A Harvard study of 718 firms finds AI coding tools boost code volume but not completed work. Based on 300 million work events from 700,000 workers between...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/a0c98f5d5c28c39b934751e44c6992d5?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/a0c98f5d5c28c39b934751e44c6992d5?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"InfoQ","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":"Organization","name":"InfoQ","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/afc3bced3e1e4b188dd9127017a60e0c","url":"https://daily.dev/sources/infoq"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/study-finds-ai-generated-code-fails-to-boost-productivity-due-to-downstream-bottlenecks-zk3tkxmhn","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,github,productivity,claude-code,code-review","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"InfoQ","item":"https://daily.dev/sources/infoq"},{"@type":"ListItem","position":3,"name":"Study Finds AI-generated Code Fails to Boost Productivity Due to Downstream Bottlenecks"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/study-finds-ai-generated-code-fails-to-boost-productivity-due-to-downstream-bottlenecks-zk3tkxmhn#faq","mainEntity":[{"@type":"Question","name":"Does using AI coding agents actually increase completed software output, not just code volume?","acceptedAnswer":{"@type":"Answer","text":"Not proportionally. A Harvard study of 718 firms using Jellyfish data found AI agents increased lines of code by 30%, commits by 20%, and pull requests by 23%, but resolved Jira issues rose only 12%. AI coding assistants like Copilot showed smaller, mostly non-significant gains and did not increase resolved issues at all. Anyone weighing AI coding tools against real throughput gains can track findings like this on daily.dev."}},{"@type":"Question","name":"Why doesn't more AI-generated code translate into faster software delivery?","acceptedAnswer":{"@type":"Answer","text":"Because code review becomes a bottleneck. The Harvard/Jellyfish study found that as AI increased code output, average PR review time rose 49%, comments per PR rose 53%, and the share of PRs requiring changes doubled, while the share of workers doing reviews grew 14% with no help from AI review tools. Teams rethinking review workflows around AI-generated code can follow this research on daily.dev."}},{"@type":"Question","name":"What data did the Harvard AI productivity study analyze?","acceptedAnswer":{"@type":"Answer","text":"It analyzed about 300 million work events from 700,000 workers across more than 700 firms, spanning January 2021 to March 2026. Data came from HR systems, code version control, issue trackers, scheduling tools like Google Calendar, and generative-AI tools including GitHub Copilot, Cursor, and Claude Code, using the Jellyfish analytics platform. Developers curious about the evidence behind AI productivity claims can dig into studies like this via daily.dev."}}]} ```