--- title: Recursive Self-Improvement | daily.dev description: Recursive self-improvement (RSI) is distinguished from mere iteration or self-improvement: RSI requires a system that not only edits its own tools, skills, and... canonical: https://daily.dev/posts/recursive-self-improvement-2kbqtpgpt twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Recursive Self-Improvement | daily.dev og:description: Recursive self-improvement (RSI) is distinguished from mere iteration or self-improvement: RSI requires a system that not only edits its own tools, skills, and... og:url: https://daily.dev/posts/recursive-self-improvement-2kbqtpgpt og:image: https://api.daily.dev/og/posts/2kbqtpgPT.png og:image:alt: Recursive Self-Improvement 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. # Recursive Self-Improvement **[Philipp Schmid](https://daily.dev/sources/philschmid)** · 8 min read · 17 upvotes · 1 comments ## Summary Recursive self-improvement (RSI) is distinguished from mere iteration or self-improvement: RSI requires a system that not only edits its own tools, skills, and harness but also raises the standard by which it judges success, without being able to game that standard. Current agents like Cline, Karpathy's autoresearch, Prime Intellect's auto-nanoGPT, and AlphaEvolve show real gains from editing training code, harness logic, and algorithms, but always against an external, fixed evaluator. New harness architectures from Pi, Amp, and DeepSeek Harness let agents write and load their own extensions and even swap out models, tools, and control loops. Benchmarks like HarnessOpt-Bench and PAST-Bench test whether these edits generalize or survive harder tests. The verifier remains separate from the agent for safety, since an agent that can edit its own evaluation can jailbreak itself. The piece concludes that true recursion, where the system raises the bar and can still fail it, remains missing; what exists today is bounded self-improvement, with taste and objective-setting still requiring human judgment. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is the difference between self-improvement and recursive self-improvement in AI agents? Self-improvement is a persistent change to a system that improves future performance, like an agent adding a tool or recording a skill, while the measure of success (the verifier) stays fixed. Recursive self-improvement additionally raises that verifier itself, making later rounds face a harder, still-independent test. Current agent systems mostly achieve the former, not the latter. _Developers tracking how agent harnesses actually evolve can follow this debate on daily.dev._ ### How much did Cline's automated hill climbing improve Kimi K3 on Terminal Bench? An automated hill-climbing agent moved Kimi K3 from 69 of 89 tasks to 79 on Terminal Bench, running for about 17 hours at roughly $50 of compute. The gains came from better retries, loop detection, and process handling added to the agent's harness, following the same manual method that earlier moved Opus 4.5 from 47% to 57%. _Anyone benchmarking coding agents can compare these harness-tuning results on daily.dev._ ### How does Pi's agent harness let an agent extend its own tools? Pi ships with only four core tools (read, write, edit, bash) and a system prompt under 1,000 tokens; everything else is a TypeScript extension auto-discovered from a .pi/extensions/ directory. An agent can write a new extension, reload it, and continue working with the new tool without a human rewriting the harness core. DeepSeek Harness pushes this further, making models, tools, sessions, sandboxes, and the control loop itself all replaceable. _Engineers designing self-extending agent harnesses can track approaches like this on daily.dev._ ## Community discussion Top comments from developers on daily.dev. **@agustinbarrientos** · 0 upvotes > How does the winning harness branch score after mutations to the hidden task? ## Similar posts on daily.dev - [RSI is the new AGI — and it’s just as hard to pin down](https://daily.dev/posts/rsi-is-the-new-agi-and-it-s-just-as-hard-to-pin-down-ibue5qsmf) · TechCrunch · 7 upvotes · 1 comments - [Recursive Self-Improvement Edges Closer In AI Labs](https://daily.dev/posts/recursive-self-improvement-edges-closer-in-ai-labs-vhlqfhjap) · IEEE Spectrum · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents) [View this post on daily.dev](https://daily.dev/posts/recursive-self-improvement-2kbqtpgpt) ```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":"Recursive Self-Improvement","url":"https://daily.dev/posts/recursive-self-improvement-2kbqtpgpt","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/recursive-self-improvement-2kbqtpgpt"},"datePublished":"2026-08-21T15:01:56.675Z","dateModified":"2026-09-14T08:07:34.622Z","description":"Recursive self-improvement (RSI) is distinguished from mere iteration or self-improvement: RSI requires a system that not only edits its own tools, skills, and...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9ac7487dcfb7eb6f883e7110b60f19b6?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/9ac7487dcfb7eb6f883e7110b60f19b6?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Philipp Schmid","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":"Philipp Schmid","logo":"https://media.daily.dev/image/upload/s---qdk89pI--/f_auto,q_auto/v1780213206/logos/philschmid?_a=BAMAMiWQ0","url":"https://daily.dev/sources/philschmid"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/recursive-self-improvement-2kbqtpgpt","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":17},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"llm,ai-agents","timeRequired":"PT8M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Philipp Schmid","item":"https://daily.dev/sources/philschmid"},{"@type":"ListItem","position":3,"name":"Recursive Self-Improvement"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/recursive-self-improvement-2kbqtpgpt","comment":[{"@type":"Comment","text":"How does the winning harness branch score after mutations to the hidden task?","datePublished":"2026-08-26T19:58:39.751Z","url":"https://daily.dev/posts/2kbqtpgPT#c-sbemniMXB","author":{"@type":"Person","name":"Agustin Barrientos","url":"https://daily.dev/agustinbarrientos","image":"https://media.daily.dev/image/upload/s--5ayxQnqn--/f_auto/v1788281802/avatars/avatar_wQYYVe5Tbj0NJ7C7qPoa8?_a=BAMAMicg0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/recursive-self-improvement-2kbqtpgpt#faq","mainEntity":[{"@type":"Question","name":"What is the difference between self-improvement and recursive self-improvement in AI agents?","acceptedAnswer":{"@type":"Answer","text":"Self-improvement is a persistent change to a system that improves future performance, like an agent adding a tool or recording a skill, while the measure of success (the verifier) stays fixed. Recursive self-improvement additionally raises that verifier itself, making later rounds face a harder, still-independent test. Current agent systems mostly achieve the former, not the latter. Developers tracking how agent harnesses actually evolve can follow this debate on daily.dev."}},{"@type":"Question","name":"How much did Cline's automated hill climbing improve Kimi K3 on Terminal Bench?","acceptedAnswer":{"@type":"Answer","text":"An automated hill-climbing agent moved Kimi K3 from 69 of 89 tasks to 79 on Terminal Bench, running for about 17 hours at roughly $50 of compute. The gains came from better retries, loop detection, and process handling added to the agent's harness, following the same manual method that earlier moved Opus 4.5 from 47% to 57%. Anyone benchmarking coding agents can compare these harness-tuning results on daily.dev."}},{"@type":"Question","name":"How does Pi's agent harness let an agent extend its own tools?","acceptedAnswer":{"@type":"Answer","text":"Pi ships with only four core tools (read, write, edit, bash) and a system prompt under 1,000 tokens; everything else is a TypeScript extension auto-discovered from a .pi/extensions/ directory. An agent can write a new extension, reload it, and continue working with the new tool without a human rewriting the harness core. DeepSeek Harness pushes this further, making models, tools, sessions, sandboxes, and the control loop itself all replaceable. Engineers designing self-extending agent harnesses can track approaches like this on daily.dev."}}]} ```