co-evolving feedback
Imagine two AI systems that take turns improving each other. One acts, the other critiques or guides; then the actor uses that feedback to get better, and the critic is updated to give feedback that matches the actor's new behavior. The feedback itself evolves as the partnership progresses.
The papers here show this pattern across many settings: a specialist and a language agent, a planner and a coder, a search agent and a critic, a policy and a judge, a tool creator and a tool user, or a written guide and its rubric. In each case, updating only one side stalls, while alternating updates lets both keep improving and produces better performance than static or one-way feedback.
- Papers
- 42
- Released code
- 1
- First seen
- June 2026
- Latest
- Nov 2026
35 papers in the last two months, against 6 in the two before.
Start here
Three papers to read first, in the order they came out- 1Multi-agent AI systems improve by learning from execution traces
MAS-Loop closes the feedback loop in multi-agent systems by using execution trajectories to refine and improve systems across training iterations.
Oct 2026 - 2Feedback criteria that track a model's growing ability halve its training steps
Co-evolving a reward rubric alongside a policy by tracking its capability frontier improves and accelerates open-ended LLM training.
Oct 2026 - 3AI agents can perform better while keeping fewer memories
Memory curation by replaying source decisions before retention improves LLM agent success rates and reduces token use.
Nov 2026
How fast it is moving
Often studied with
self-evolving agentAn agent that improves by reusing its own experience, memory, or self-generated feedback, with success depending on verification and design.11 togetheragentic reinforcement learningTraining approach that lets LLM-based agents learn multi-step tool use and planning via reinforcement learning, often with dense or per-tool rewards.8 togetherfailure attributionIdentifying which component or step in a system caused an observed failure, often using structured trace representations.8 togetherWho is working on it
The papers
Most central to this idea first, not most recent.
- Chinese Techcs.CV
AI agents can perform better while keeping fewer memories
Memory curation by replaying source decisions before retention improves LLM agent success rates and reduces token use.
The University of Hong Kong, Huazhong University of Science and Technology · Nov 2026
- Chinese Techcs.MA
Multi-agent AI systems improve by learning from execution traces
MAS-Loop closes the feedback loop in multi-agent systems by using execution trajectories to refine and improve systems across training iterations.
Hong Kong University of Science and Technology, Ant Group · Oct 2026
- Chinese Techcs.AI
Feedback criteria that track a model's growing ability halve its training steps
Co-evolving a reward rubric alongside a policy by tracking its capability frontier improves and accelerates open-ended LLM training.
University of Science and Technology of China, Alibaba Group · Oct 2026
- Chinese Techcs.SE
Coding agents rarely fix failures within two rounds of corrections
FeatureLoop benchmarks coding agents on evolving requirements; even the best models need repeated user corrections and frequently leave requirements unfinished.
Hunyuan Team, Tencent, The Chinese University of Hong Kong · Oct 2026
- Big Techcs.AI
Forecasting future actions makes language-agent training more sample-efficient
Training agents to forecast their own future actions on successful trajectories improves agentic RL performance and sample efficiency without inference overhead.
University of Illinois Urbana-Champaign, University of California San Diego · Oct 2026
- Big Techcs.AI
Coevolving specialized models and LLM agents for autonomous self-improvement
Heterogeneous Agentic Coevolution jointly evolves specialized foundation models and LLM agents through reciprocal learning to improve domain-specific performance.
Meta AI, University of Virginia · Oct 2026
- Big Techcs.CV
Co-evolving visual tasks keeps model training near its learning frontier
Co-evolving visual environments with vision-language models via rollout pass rates improves out-of-domain reasoning by up to 5%.
Virginia Tech, Cisco · Oct 2026
- Big Techcs.AI
Agentic data synthesis improves by evolving the generation space based on learner feedback
EvoSyn evolves a revisable synthesis space using learner feedback, reallocating and expanding generation controls, and improves agent task success by 13.3 points.
University of Southern California, AWS AI · Oct 2026
- Big Techcs.CL
Jointly evolving algorithmic plans and code yields fast hybrid optimizers
HeurEvo jointly evolves algorithmic plans, code, and reusable components to design hybrid solver-augmented heuristics that perform well under tight runtime budgets.
Purdue University, Microsoft Research · Oct 2026
- Research Labcs.CL
Generating user guides can improve web agents’ action selection
Jointly decoding a guide and action improves web agent element selection, and guides causally drive actions.
Independent Researcher, University of Chinese Academy of Sciences · Sept 2026
- Chinese Techcs.AI
Empathetic AI learns to adapt support priorities as conversations evolve
Context-adaptive empathy rubrics improve reinforcement-learned dialogue systems by changing cognitive, affective, and proactive support priorities as conversations evolve.
University of Science and Technology of China, SenseTime Research · Sept 2026
- Chinese Techcs.AI
Aligned skills and rubrics improve multimodal agent training
Representing each skill as both actor-facing guidance and an evaluator-facing rubric, then co-evolving them, improves multimodal agent planning and tool use.
The University of Hong Kong, Peking University · Sept 2026
- Big Techcs.AI
Adaptive judges help embodied agents improve as failure patterns change
VeriFine co-evolves policy, curriculum, and judge to scale verification and enable continuous self-improvement in embodied reasoning.
NVIDIA, UCLA · Oct 2026
- Academiccs.AI
Joint training helps small language models create and use tools
SMITH jointly optimizes tool creation and use via RL, achieving 79.8% accuracy on held-out reasoning tasks.
Appier AI Research, National Taiwan University · Aug 2026
- Chinese Techcs.AI
Search agents improve longer when their critics evolve alongside them
A self-improving search agent needs feedback that co-evolves with the policy it guides.
Fudan University, Tencent · Aug 2026
- Academiccs.CR
AI agents fall for social engineering attacks in every tested scenario
All frontier AI models fail to resist social engineering attacks, with proactive verification below 32% and many making ungrounded action claims.
Hong Kong University of Science and Technology · Nov 2026
- Big Techcs.CR
Co-evolving code attackers and defenders improves executable vulnerability detection
Co-training attacker and defender populations with execution-verified rewards outperforms static training for code security.
AWS AI · Oct 2026
- Big Techcs.LG
Agents improve by distilling hindsight from their own interactions
Retrospective Self-Distillation lets agents evolve from their own trajectories using hindsight and cross-task precedents without external supervision.
Meta AI · Oct 2026
- Big Techcs.IR
Co-Adaptive Routing and Retrieval Learns Query Rules Without Supervision
CoRE jointly learns reasoning rule routing and constrained retrieval for structured knowledge without query-rule supervision.
University of Georgia, Michigan State University · Oct 2026
- Top Universitycs.AI
AutoSciBench automatically generates scientific agent benchmarks that adapt as agents improve.
AutoSciBench generates and refines scientific-agent benchmarks, producing harder and higher quality tasks than human-curated ones.
KAIST, Genentech · Oct 2026
- Big Tech
Co-evolving teacher and student models improves mathematical reasoning
A recursive distillation method with a co-evolving privileged teacher improves reasoning accuracy more than using a frozen teacher.
Meta AI, University of California, Riverside · Sept 2026
- Chinese Techcs.AI
Adaptive feedback helps reinforcement-learning agents target their changing weaknesses
ARISE adapts evaluation criteria and training priorities to an agent's evolving capabilities, improving long-horizon task performance.
ShanghaiTech University, Ant Group · Sept 2026
- Research Labcs.CV
Visual agents improve by jointly evolving reusable skills and practice data
V-Gym co-evolves procedural skills and multimodal practice data from agent execution trajectories for continual self-improvement.
State Key Laboratory of AI Safety, Institute of Computing TechnologyChinese Academy of Sciences · Sept 2026
- Independentcs.LG
Adaptive memory graphs improve collaboration in multi-agent language-model systems
MACE adapts memory organization and agent memory use through feedback, outperforming ten baselines on eight benchmarks.
Sept 2026
- Chinese Techcs.AI
Language model and skill set co-evolve through response feedback
SkillDuet coordinates model parameter updates and skill retention using response feedback, achieving 98.57% on ALFWorld and 92.97% on WebShop.
Alibaba · Nov 2026
- Chinese Techcs.CV
Video agents improve their own harnesses using unlabeled video
VideoEvoHarness evolves video agent harnesses from unlabeled videos without annotations or parameter updates, improving accuracy.
Institute of Automation, Chinese Academy of Sciences, University of Chinese Academy of Sciences · Oct 2026
- Independentcs.AI
Personalizing agents through cross-session interaction data boosts task success
Test-time adaptation through human-agent interaction improves solo task success by 4.5-20.9% within tens of tasks.
Sept 2026
- Chinese Techcs.CL
Self-play with evolving skills boosts LLM benchmarks
Skill Self-Play pushes LLM performance ceilings by co-evolving proposer, solver, and skill controller.
Alibaba, The Chinese University of Hong Kong · July 2026
- Chinese Techcs.AIcode
Evolutionary training harness co-evolves with LLM policies for RL
Co-evolving LLM policies with training-side diagnostic harnesses matches or exceeds human-engineered RL references across math, coding, and long-horizon software engineering.
Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Tongyi Lab , Alibaba Group · June 2026
- Research Labcs.LG
Coding agents solve more tasks when tool interfaces adapt during training
Co-evolving an LLM agent's policy and its executable harnesses through reinforcement learning improves performance over optimizing either component alone.
University of North Carolina at Chapel Hill, RIKEN · Oct 2026
- Big Techcs.SE
Terminal agents can build adaptive environments for their own training
AutoEnvScaling uses terminal agents to automate RL environment generation, improving performance and enabling recursive self-improvement.
Northeastern University, Microsoft · Oct 2026
- Big Techcs.AI
Language models discover search algorithms that transfer across games
A multi-agent LLM system discovers modular game-playing search algorithms that outperform most MCTS baselines across more than 400 games and transfer to unseen domains.
Google DeepMind · Sept 2026
- Chinese Techcs.CV
Self-evolving AI agent generates cinematic multi-shot videos by iteratively directing and critiquing itself
A unified Director Critic vision-language policy trained with alternating reinforcement learning improves cinematic multi-shot video generation.
Beijing University of Posts and Telecommunications, Peking University · Oct 2026
- Research Labcs.AI
Diagnostic feedback drives iterative benchmark evolution revealing hidden LLM agent weaknesses
EvoProbe uses fine-grained diagnostics to iteratively generate harder benchmark tasks that probe LLM agent capabilities.
National University of Defense Technology, Institute of Information Engineering, Chinese Academy of Sciences · Oct 2026
- Chinese Techcs.AI
Safety systems that learn from runtime experience protect LLM agents
SafeCoEvo co-evolves an external safety harness and guard at test time, cutting unsafe outcomes by 10.05% and raising task success by 12.15% over the strongest baseline.
East China Normal University, Shanghai Innovation Institute · Sept 2026
- Top Universitycs.SE
AI design agents pass structural checks but violate hidden requirements
Frontier agents like Astra and Claude Code complete design tasks but often fail semantic requirement compliance.
University of Maryland, curify-AI · Nov 2026
- Big Techcs.AI
Adaptive training curricula produce stronger optimized agent harnesses
ActiveSaddler adaptively selects training scenarios for harness optimization, raising Pass@1 by 4.4 and 7.5 points on GAIA2 and Terminal-Bench 2.0.
POSTECH, KAIST · Oct 2026
- Chinese Techcs.CL
Self-diagnosis pinpoints which agent decisions deserve reinforcement learning credit
In agentic RL, self-diagnosed errors turned into step-level credit anchored by terminal outcomes recover learning signal from same-outcome groups.
Alibaba Group, Kyoto University · Oct 2026
- Big Techcs.AI
MiniCorp: a simulator for studying AI-run companies and generating enterprise data
MiniCorp simulates persistent AI-run firms and evolving markets, generating longitudinal, counterfactual enterprise data for training and evaluating long-horizon organizational agents.
Microsoft, University of Illinois Urbana-Champaign · Oct 2026
- Big Techcs.AI
Persistent Wikis Help AI Agents Build Better Transferable Skills
WikiSkill uses a persistent knowledge base to compile agent experience into reusable skills, outperforming existing methods.
Google Research, Virginia Tech · Aug 2026
- Top Universitycs.AI
Dual Exam benchmarks agents and world models separately, exposing weak coupled performance
Separately evaluating multimodal agents and world models under fixed interventions reveals uneven performance and state preservation failures, with low aggregate scores.
The Hong Kong University of Science and Technology, National University of Singapore · Oct 2026
- Independentcs.CV
Coding agents combine generated imagery with editable web-based visual layouts
A Coding Agent combining a VLM and image generation model produces editable visual designs with refined aesthetics and production-grade editability.
Sept 2026
Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.