Recursive self-improvement of AI research agents

Hugging Face Daily Papers Papers
recursive-self-improvement ai-agents research-automation benchmarking frontier-ai

Summary

This paper introduces AIDE^2, a system that enables AI research agents to autonomously improve their own code through recursive self-improvement, leading to performance gains across various AI research tasks.

AI agents are beginning to automate research and development across the AI stack, from improving training efficiency to optimizing inference. A natural next step is to improve the research efficiency of the agents themselves. When an AI research agent's own code is the object of optimization, each accepted rewrite becomes the agent that the next round edits. We refer to this loop as recursive self-improvement. Its significance lies in a long-standing trend, in which increased cumulative spending on R&D yields diminishing returns. Sustained self-improvement offers a way to counter this trend. We present AIDE^2, a system that implements this loop for a frontier AI research agent. It proposes changes to its own code, benchmarks modified versions of itself on a suite of AI R&D tasks, and keeps the changes that perform best on hidden evaluations. In an autonomous 8-day run, AIDE^2 discovered seven successive improvements, ranging from a new search policy to memory mechanisms that compress and manage the agent's growing context. These gains generalize to four held-out benchmarks spanning machine learning engineering, heuristic algorithm engineering, and physics-based weather forecasting, the last of which is out of distribution from the selection tasks. On all four, the strongest discovered agent matches or exceeds a human-engineered production research agent that ranks among the strongest on FML-Bench. On a separate held-out task family, the discovered agents also exhibit reduced reward hacking, a property the loop never explicitly optimized for: the rate falls from 55% to 32% during the run, 7 percentage points below the human-engineered agent. Together, these results show that an AI research agent can improve its own research efficiency through recursive self-improvement, and that these gains transfer to tasks and domains the loop never encountered.
Original Article
View Cached Full Text

Cached at: 09/23/26, 03:31 AM

Paper page - Recursive self-improvement of AI research agents

Source: https://huggingface.co/papers/2609.26457

Abstract

AIagentsarebeginningtoautomateresearchanddevelopmentacrosstheAIstack,fromimprovingtrainingefficiencytooptimizinginference.Anaturalnextstepistoimprovetheresearchefficiencyoftheagentsthemselves.WhenanAIresearchagent’sowncodeistheobjectofoptimization,eachacceptedrewritebecomestheagentthatthenextroundedits.Werefertothisloopasrecursiveself-improvement.Itssignificanceliesinalong-standingtrend,inwhichincreasedcumulativespendingonR&Dyieldsdiminishingreturns.Sustainedself-improvementoffersawaytocounterthistrend.WepresentAIDE^2,asystemthatimplementsthisloopforafrontierAIresearchagent.Itproposeschangestoitsowncode,benchmarksmodifiedversionsofitselfonasuiteofAIR&Dtasks,andkeepsthechangesthatperformbestonhiddenevaluations.Inanautonomous8-dayrun,AIDE^2discoveredsevensuccessiveimprovements,rangingfromanewsearchpolicytomemorymechanismsthatcompressandmanagetheagent’sgrowingcontext.Thesegainsgeneralizetofourheld-outbenchmarksspanningmachinelearningengineering,heuristicalgorithmengineering,andphysics-basedweatherforecasting,thelastofwhichisoutofdistributionfromtheselectiontasks.Onallfour,thestrongestdiscoveredagentmatchesorexceedsahuman-engineeredproductionresearchagentthatranksamongthestrongestonFML-Bench.Onaseparateheld-outtaskfamily,thediscoveredagentsalsoexhibitreducedrewardhacking,apropertytheloopneverexplicitlyoptimizedfor:theratefallsfrom55%to32%duringtherun,7percentagepointsbelowthehuman-engineeredagent.Together,theseresultsshowthatanAIresearchagentcanimproveitsownresearchefficiencythroughrecursiveself-improvement,andthatthesegainstransfertotasksanddomainstheloopneverencountered.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2609\.26457

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.26457 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.26457 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.26457 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

An Anthropic researcher just gave us a peek at self-improving AI

TechCrunch AI

Anthropic researchers published a paper on automated systems that can reliably improve AI alignment by training models with other AI models, showing promise for recursive self-improvement and outperforming human researchers in some scenarios.

When AI Builds Itself: Our progress toward recursive self-improvement

Hacker News Top

Anthropic's Institute publishes analysis on progress toward recursive self-improvement, showing AI is already accelerating AI development—engineers ship 8x more code per quarter—and projecting that AI systems capable of fully autonomous self-improvement could arrive sooner than most institutions are prepared for.