--- title: Benchmarking Coding Agents on New vs Legacy Codebases —... description: A talk by Denys Linkov of Wisedocs describing a 6-month codebase refactor that consolidated 10+ legacy repos into a monorepo, driven by scaling pain and AI... canonical: https://daily.dev/posts/benchmarking-coding-agents-on-new-vs-legacy-codebases-denys-linkov-wisedocs-z0xzd0s9r twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Benchmarking Coding Agents on New vs Legacy Codebases — Denys Linkov, Wisedocs | daily.dev og:description: A talk by Denys Linkov of Wisedocs describing a 6-month codebase refactor that consolidated 10+ legacy repos into a monorepo, driven by scaling pain and AI... og:url: https://daily.dev/posts/benchmarking-coding-agents-on-new-vs-legacy-codebases-denys-linkov-wisedocs-z0xzd0s9r og:image: https://api.daily.dev/og/posts/Z0XzD0S9r.png og:image:alt: Benchmarking Coding Agents on New vs Legacy Codebases — Denys Linkov, Wisedocs 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. # Benchmarking Coding Agents on New vs Legacy Codebases — Denys Linkov, Wisedocs **[AI Engineer](https://daily.dev/sources/aidotengineer)** · 18 min read · 0 upvotes · 0 comments ## Summary A talk by Denys Linkov of Wisedocs describing a 6-month codebase refactor that consolidated 10+ legacy repos into a monorepo, driven by scaling pain and AI pipeline complexity. The talk benchmarks coding agents (o3, Claude Sonnet, Opus) on real refactoring tasks, showing dramatic improvement over time — a task that took 3 hours with 10 mistakes using o3 can now be near-one-shotted with Opus. Key findings: shipping velocity increased dramatically post-refactor, developer participation grew, and features that took months now ship in under a week. The speaker also argues that measuring AI agent reliability at 90–99% accuracy is more useful than the commonly cited 50%, and warns against 'AI psychosis' — mistaking impressive-looking AI output for correct output. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much faster did newer AI models complete the same code refactoring task compared to older ones? A refactoring task that took OpenAI's o3 three hours of back-and-forth in Cursor and still produced 10 major mistakes was solved by Claude Sonnet 4.5 with one extra iteration, and Claude Opus essentially one-shot it. Rebuilding the same task with modern models and harnesses was estimated to take roughly a fifth of the original time. _daily.dev helps developers track how fast AI coding agents are actually improving on real refactoring work._ ### Can a large language model fully refactor a legacy codebase into a monorepo on its own? Not reliably yet. An attempt to have GPT-5.5 zero-shot a full codebase refactor completed in about 10 minutes but only produced roughly 2,000 lines of scaffolding code without implementing the actual models or a working deployment, meaning it looked done but wasn't functional. _Engineers weighing how much to trust autonomous refactors can follow these agent capability gaps on daily.dev._ ### Is it worth refactoring a legacy AI pipeline now or waiting for AI coding models to improve further? Refactoring now was judged worthwhile: a six-month effort to consolidate more than 10 repositories into a monorepo reduced pipeline processing time and costs, supported larger files, cut feature delivery from multiple months to under a week, and made more developers willing to contribute to the cleaner codebase. _Teams weighing refactor timing against improving AI tooling can compare real outcomes like this on daily.dev._ ## Similar posts on daily.dev - [Agents Refactor 300K Lines in Three Weeks, and Practitioners Ask What It Proves](https://daily.dev/posts/agents-refactor-300k-lines-in-three-weeks-and-practitioners-ask-what-it-proves-qtk43gheh) · InfoQ · 0 upvotes · 0 comments - [When Code Becomes Cheap, What’s Left?](https://daily.dev/posts/when-code-becomes-cheap-what-s-left--jbyzzm8t6) · Atomic Spin · 2 upvotes · 0 comments - [The 80% Problem in Agentic Coding](https://daily.dev/posts/the-80-problem-in-agentic-coding-9nacygamm) · Elevate · 120 upvotes · 7 comments - [I benchmarked Claude Code and OpenCode on a heavy refactor: The reality of agentic CLI workflows](https://daily.dev/posts/i-benchmarked-claude-code-and-opencode-on-a-heavy-refactor-the-reality-of-agentic-cli-workflows-ohlo6vfwb) · LogRocket · 1 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#general-programming](https://daily.dev/tags/general-programming), [#ai-agents](https://daily.dev/tags/ai-agents), [#technical-debt](https://daily.dev/tags/technical-debt) [View this post on daily.dev](https://daily.dev/posts/benchmarking-coding-agents-on-new-vs-legacy-codebases-denys-linkov-wisedocs-z0xzd0s9r) ```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":"Benchmarking Coding Agents on New vs Legacy Codebases — Denys Linkov, Wisedocs","url":"https://daily.dev/posts/benchmarking-coding-agents-on-new-vs-legacy-codebases-denys-linkov-wisedocs-z0xzd0s9r","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/benchmarking-coding-agents-on-new-vs-legacy-codebases-denys-linkov-wisedocs-z0xzd0s9r"},"datePublished":"2026-08-08T19:06:56.116Z","dateModified":"2026-09-13T21:45:41.048Z","description":"A talk by Denys Linkov of Wisedocs describing a 6-month codebase refactor that consolidated 10+ legacy repos into a monorepo, driven by scaling pain and AI...","image":"https://i.ytimg.com/vi/7vn4WpqNpck/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/7vn4WpqNpck/sddefault.jpg","isAccessibleForFree":true,"articleSection":"AI Engineer","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":"AI Engineer","logo":"https://media.daily.dev/image/upload/s--u5PucxNT--/f_auto/v1724338940/logos/aidotengineer","url":"https://daily.dev/sources/aidotengineer"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/benchmarking-coding-agents-on-new-vs-legacy-codebases-denys-linkov-wisedocs-z0xzd0s9r","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai,general-programming,ai-agents,technical-debt","timeRequired":"PT18M","video":{"@type":"VideoObject","name":"Benchmarking Coding Agents on New vs Legacy Codebases — Denys Linkov, Wisedocs","description":"A talk by Denys Linkov of Wisedocs describing a 6-month codebase refactor that consolidated 10+ legacy repos into a monorepo, driven by scaling pain and AI...","thumbnailUrl":"https://i.ytimg.com/vi/7vn4WpqNpck/sddefault.jpg","uploadDate":"2026-08-08T19:06:56.116Z","duration":"PT18M","url":"https://api.daily.dev/r/Z0XzD0S9r","embedUrl":"https://www.youtube.com/embed/7vn4WpqNpck"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"AI Engineer","item":"https://daily.dev/sources/aidotengineer"},{"@type":"ListItem","position":3,"name":"Benchmarking Coding Agents on New vs Legacy Codebases — Denys Linkov, Wisedocs"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/benchmarking-coding-agents-on-new-vs-legacy-codebases-denys-linkov-wisedocs-z0xzd0s9r#faq","mainEntity":[{"@type":"Question","name":"How much faster did newer AI models complete the same code refactoring task compared to older ones?","acceptedAnswer":{"@type":"Answer","text":"A refactoring task that took OpenAI's o3 three hours of back-and-forth in Cursor and still produced 10 major mistakes was solved by Claude Sonnet 4.5 with one extra iteration, and Claude Opus essentially one-shot it. Rebuilding the same task with modern models and harnesses was estimated to take roughly a fifth of the original time. daily.dev helps developers track how fast AI coding agents are actually improving on real refactoring work."}},{"@type":"Question","name":"Can a large language model fully refactor a legacy codebase into a monorepo on its own?","acceptedAnswer":{"@type":"Answer","text":"Not reliably yet. An attempt to have GPT-5.5 zero-shot a full codebase refactor completed in about 10 minutes but only produced roughly 2,000 lines of scaffolding code without implementing the actual models or a working deployment, meaning it looked done but wasn't functional. Engineers weighing how much to trust autonomous refactors can follow these agent capability gaps on daily.dev."}},{"@type":"Question","name":"Is it worth refactoring a legacy AI pipeline now or waiting for AI coding models to improve further?","acceptedAnswer":{"@type":"Answer","text":"Refactoring now was judged worthwhile: a six-month effort to consolidate more than 10 repositories into a monorepo reduced pipeline processing time and costs, supported larger files, cut feature delivery from multiple months to under a week, and made more developers willing to contribute to the cleaner codebase. Teams weighing refactor timing against improving AI tooling can compare real outcomes like this on daily.dev."}}]} ```