--- title: Five studies that are changing how I think about AI in... description: A synthesis of five recent research papers examining AI's impact on software engineering. Key findings: GitHub Copilot shows ~40% more PR throughput in... canonical: https://daily.dev/posts/five-studies-that-are-changing-how-i-think-about-ai-in-software-engineering-oltg7iioa twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Five studies that are changing how I think about AI in software engineering | daily.dev og:description: A synthesis of five recent research papers examining AI's impact on software engineering. Key findings: GitHub Copilot shows ~40% more PR throughput in... og:url: https://daily.dev/posts/five-studies-that-are-changing-how-i-think-about-ai-in-software-engineering-oltg7iioa og:image: https://api.daily.dev/og/posts/OlTg7iioa.png og:image:alt: Five studies that are changing how I think about AI in software engineering 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. # Five studies that are changing how I think about AI in software engineering **[Engineering Enablement](https://daily.dev/sources/engineering_enablement)** · 11 min read · 3 upvotes · 0 comments ## Summary A synthesis of five recent research papers examining AI's impact on software engineering. Key findings: GitHub Copilot shows ~40% more PR throughput in dose-response analysis; productivity gains from AI coding tools attenuate significantly through the delivery pipeline (large coding gains shrink to ~30% more releases shipped); a longitudinal study reveals a 'productivity-experience paradox' where developer experience degrades even as throughput improves; developers want AI focused on verification and downstream tasks, not more code generation; and a conceptual paper argues technical debt is being joined by 'cognitive debt' (eroded team understanding) and 'intent debt' (lost rationale and goals) as the more critical software health concerns in an AI-assisted world. The overarching conclusion: AI is compressing upstream code generation faster than organizations are adapting their review, verification, and understanding processes downstream. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much more productive are developers who use GitHub Copilot heavily compared to weeks with no usage? Weeks with the highest Copilot usage were associated with roughly 40% more completed pull requests per hour of active coding time compared to weeks with no usage, based on an observational study tracking 16,223 developers over 43 weeks. The effect showed a dose-response pattern and was strongest for larger PRs touching seven or more files, not just small ones, after seven robustness tests ruled out alternative explanations. _daily.dev surfaces research like this for teams weighing how much AI coding tools actually move the needle._ ### Do AI coding productivity gains actually translate into more shipped software? Not proportionally. An NBER analysis of over 100,000 GitHub developers found commit increases ranging from about 40% with autocomplete tools up to 180% with autonomous coding agents, but the effect on actual shipped releases topped out at roughly 30% more releases. The estimated elasticity of substitution between AI output and human effort is low (about 0.25), meaning AI and humans remain complements, not substitutes, since review, integration, and validation still require substantial human effort. _Teams deciding how to measure AI coding ROI can track this kind of delivery-pipeline research on daily.dev._ ### Does using AI coding assistants improve or worsen developer experience over time? Both, and they appear to be decoupling. A six-month longitudinal study of 95 professional software engineers found 84% reported productivity improvement at both survey points, but the share reporting worse developer experience on at least one dimension nearly doubled, from 14% to 27%. Flow state eroded the most, cognitive load eroded modestly, while feedback loops actually improved, and productivity change scores no longer correlated with developer experience change scores. _Engineers tracking the tradeoffs of AI-assisted workflows can follow developer experience research on daily.dev._ ## Similar posts on daily.dev - [The current impact of AI on engineering velocity](https://daily.dev/posts/the-current-impact-of-ai-on-engineering-velocity-lic0nalsa) · Software Engineering Research · 0 upvotes · 0 comments - [8 Myths on Software Engineering and AI](https://daily.dev/posts/8-myths-on-software-engineering-and-ai-jfo8wdypp) · Engineering Enablement · 36 upvotes · 2 comments - [Survey Sees AI Driving DevOps Productivity Gains Despite Challenges](https://daily.dev/posts/survey-sees-ai-driving-devops-productivity-gains-despite-challenges-xuh0ydsgz) · DevOps.com · 0 upvotes · 0 comments - [How AI coding makes developers 56% faster and 19% slower](https://daily.dev/posts/how-ai-coding-makes-developers-56-faster-and-19-slower-t4jx0ry0u) · The New Stack · 0 upvotes · 0 comments --- Tags: [#github](https://daily.dev/tags/github), [#productivity](https://daily.dev/tags/productivity), [#ai-coding](https://daily.dev/tags/ai-coding), [#technical-debt](https://daily.dev/tags/technical-debt) [View this post on daily.dev](https://daily.dev/posts/five-studies-that-are-changing-how-i-think-about-ai-in-software-engineering-oltg7iioa) ```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":"Five studies that are changing how I think about AI in software engineering","url":"https://daily.dev/posts/five-studies-that-are-changing-how-i-think-about-ai-in-software-engineering-oltg7iioa","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/five-studies-that-are-changing-how-i-think-about-ai-in-software-engineering-oltg7iioa"},"datePublished":"2026-07-10T13:07:29.225Z","dateModified":"2026-09-14T07:36:44.244Z","description":"A synthesis of five recent research papers examining AI's impact on software engineering. Key findings: GitHub Copilot shows ~40% more PR throughput in...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/cf5094260a4251a59e16fef89d238feb?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/cf5094260a4251a59e16fef89d238feb?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Engineering Enablement","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":"Engineering Enablement","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/3c8216b036d34c1cbe1678385355f35d","url":"https://daily.dev/sources/engineering_enablement"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/five-studies-that-are-changing-how-i-think-about-ai-in-software-engineering-oltg7iioa","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"github,productivity,ai-coding,technical-debt","timeRequired":"PT11M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Engineering Enablement","item":"https://daily.dev/sources/engineering_enablement"},{"@type":"ListItem","position":3,"name":"Five studies that are changing how I think about AI in software engineering"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/five-studies-that-are-changing-how-i-think-about-ai-in-software-engineering-oltg7iioa#faq","mainEntity":[{"@type":"Question","name":"How much more productive are developers who use GitHub Copilot heavily compared to weeks with no usage?","acceptedAnswer":{"@type":"Answer","text":"Weeks with the highest Copilot usage were associated with roughly 40% more completed pull requests per hour of active coding time compared to weeks with no usage, based on an observational study tracking 16,223 developers over 43 weeks. The effect showed a dose-response pattern and was strongest for larger PRs touching seven or more files, not just small ones, after seven robustness tests ruled out alternative explanations. daily.dev surfaces research like this for teams weighing how much AI coding tools actually move the needle."}},{"@type":"Question","name":"Do AI coding productivity gains actually translate into more shipped software?","acceptedAnswer":{"@type":"Answer","text":"Not proportionally. An NBER analysis of over 100,000 GitHub developers found commit increases ranging from about 40% with autocomplete tools up to 180% with autonomous coding agents, but the effect on actual shipped releases topped out at roughly 30% more releases. The estimated elasticity of substitution between AI output and human effort is low (about 0.25), meaning AI and humans remain complements, not substitutes, since review, integration, and validation still require substantial human effort. Teams deciding how to measure AI coding ROI can track this kind of delivery-pipeline research on daily.dev."}},{"@type":"Question","name":"Does using AI coding assistants improve or worsen developer experience over time?","acceptedAnswer":{"@type":"Answer","text":"Both, and they appear to be decoupling. A six-month longitudinal study of 95 professional software engineers found 84% reported productivity improvement at both survey points, but the share reporting worse developer experience on at least one dimension nearly doubled, from 14% to 27%. Flow state eroded the most, cognitive load eroded modestly, while feedback loops actually improved, and productivity change scores no longer correlated with developer experience change scores. Engineers tracking the tradeoffs of AI-assisted workflows can follow developer experience research on daily.dev."}}]} ```