Awesome Reliable Self-Evolving Agents

A curated collection of research on self-evolving agents, advancing reliable AI self-improvement.

🚀 Project • 🧭 Taxonomy • 🗂️ Browse • 🤝 Contribute

🤝 Contributions are welcome: correct a manuscript-used record, or add the paper to the manuscript before proposing it here.

✉️ Contact: wkqscut@gmail.com, wenjinhou@zju.edu.cn, yanyuchen@zju.edu.cn, hehefan@zju.edu.cn

Five levels of self-evolution, drawn as divers descending from the surface to the sea floor

📌 Citation

If this survey is useful to your work, please cite it:

@article{wang2026diving,
  title   = {Diving into Reliable Self-Evolving Agents: A Survey},
  author  = {Kaiqi Wang and Wenjin Hou and Yuchen Yan and
             Hongrui Jia and Zhisheng Zhong and Botao Ren and
             Yifei Chen and Songyang Zhang and Yongliang Shen and
             Jun Xiao and Yi Yang and Yueting Zhuang and Hehe Fan},
  journal = {OpenReview.net},
  year    = {2026},
  url     = {https://openreview.net/forum?id=CGO1hDTHNe}
}

🧭 Why This List Is Different _{↑ contents}

The catalog follows the survey's two organizing questions: what changes during self-evolution, and what evidence can support claims of improvement?

Each transition is classified by the deepest evolution target whose active semantic change affects a decision-relevant output, update, or judgment—not by its algorithm name, training stage, or runtime components.

Deepest active evolution target	Characteristic failure	Works
Current output or task-local trajectory	Self-confirmation	42
Trainable model or policy state	Model collapse	137
Scaffold	Scaffold overfitting	257
Improver	Metric capture	21
Criterion	Criterion drift	30

L0 is task-local; L1–L4 require a retained change that affects later independent tasks or future updates. The levels describe how far a change reaches, not how capable or reliable the system is.

Under the survey's structural definition, recursive self-improvement (RSI) begins at L3 and extends at L4. This boundary does not itself establish improvement or imply accelerating gains.

Across all levels, reliable self-evolution depends on whether evaluation and oversight remain independent of the update and adequately cover the scope of the improvement claim.

Figure 1. The self-evolution loop. The agent runs a task, then proposes and selects a candidate change, and an external audit either accepts it, rejects it and rolls back, or escalates to a human. The evidence source and the acceptance gate stay outside the update boundary, so the loop cannot rewrite them.

Figure 2. What the companion survey covers. Part I frames self-evolution and RSI, Part II maps methods from L0 to L4 by evolution depth, and Part III analyses reliability and open problems.

This repository is the static companion list for the survey. The complete L0–L4 taxonomy and all 549 manuscript-used papers are presented below.

🗂️ Contents

Browse 549 works by self-evolution level

📚 Surveys and Positioning 24
✍️ L0: Output-Level Self-Evolution 42
🧠 L1: Model-Level Self-Evolution 137
🧰 L2: Scaffold-Level Self-Evolution 257
🔁 L3: Improver-Level Self-Evolution 21
🎯 L4: Criterion-Level Self-Evolution 30
🛡️ Cross-Level Reliability: Evidence, Acceptance, and Control 8
🚀 Open Problems and Outlook 30
⚖️ License

📚 Surveys and Positioning _{↑ contents}

Surveys and adjacent literature used to position the field; these works are not assigned an L0-L4 self-evolution level.

↪️ Jump to: Field Positioning and Related Surveys (24)

Field Positioning and Related Surveys

TMLR 2026 A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence. [paper] [companion]
SSRN 2026 A Systematic Survey of Self-Evolving Agents: From Model-Centric to Environment-Driven Co-Evolution. [paper]
arXiv 2026 Agentic Environment Engineering for LLMs: A Survey of Environment Modeling, Synthesis, Evaluation, and Application. [paper]
XYZ Lab 2026 AI4AI at Scale: A Full-Pipeline System for Enhancing LLM Agentic Capabilities. [paper]
arXiv 2026 Beyond Individual Intelligence: Surveying Collaboration, Failure Attribution, and Self-Evolution in LLM-based Multi-Agent Systems. [paper] [companion]
Lilian Weng Blog 2026 Harness Engineering for Self-Improvement. [paper] [project]
arXiv 2026 Memory for Autonomous LLM Agents: Mechanisms, Evaluation, and Emerging Frontiers. [paper]
arXiv 2026 Self-improvements in modern agentic systems: a survey. [paper] [code] [project]
arXiv 2026 Self-Improving Agents in the Era of Experience: A Survey of Self- to Meta-Evolution. [paper]
TMLR 2026 The Landscape of Agentic Reinforcement Learning for LLMs: A Survey. [paper] [project]
OpenReview 2026 Towards long-horizon agents: a survey. [paper]
Academia AI and Applications 2026 Towards Trustworthy Agentic AI: A Comprehensive Survey of Safety, Robustness, Privacy, and System Security. [paper]
arXiv 2025 A Comprehensive Survey of Self-Evolving AI Agents: A New Paradigm Bridging Foundation Models and Lifelong Agentic Systems. [paper] [code]
arXiv 2025 A Survey on the Memory Mechanism of Large Language Model based Agents. [paper] [companion]
Nature 2025 Optimizing Generative AI by Backpropagating Language Model Feedback. [paper]
EMNLP 2025 Self-Improvement in Multimodal Large Language Models: A Survey. [paper]
arXiv 2025 Towards Lifelong Learning of Large Language Models: A Survey. [paper] [companion]
Google DeepMind 2025 Welcome to the Era of Experience. [paper]
arXiv 2024 A Survey on LLM Inference-Time Self-Improvement. [paper] [companion]
arXiv 2024 A Survey on Self-Evolution of Large Language Models. [paper] [companion]
arXiv 2024 Internal Consistency and Self-Feedback in Large Language Models: A Survey. [paper] [companion]
ICLR 2024 PromptAgent: Strategic Planning with Language Models Enables Expert-level Prompt Optimization. [paper]
NeurIPS 2023 AdaPlanner: Adaptive Planning from Feedback with Language Models. [paper]
NeurIPS 2023 Reflexion: Language Agents with Verbal Reinforcement Learning. [paper]

✍️ L0: Output-Level Self-Evolution _{↑ contents}

Deepest active evolution target: Current output or task-local trajectory. Characteristic failure: Self-confirmation.

Task-local workflow of Output-Level Self-Evolution

Figure 3. Reflection, exploration, and verification revise the current output while the underlying agent setup stays fixed, so independent tasks start fresh.

↪️ Jump to: Task-Local Boundary and Persistence (1) · Iterative Revision (11) · Search, Verification, and Acceptance (28) · Reliability and the Persistence Limit (2)

Task-Local Boundary and Persistence

arXiv 2026 Active Context Compression: Autonomous Memory Management in LLM Agents. [paper]

Iterative Revision

arXiv 2026 Kestrel: Grounding Self-Refinement for LVLM Hallucination Mitigation. [paper]
ECCV 2026 Reflect-R1: Evidence-Driven Reflection for Self-Correction in Long Video Understanding. [paper] [code]
PACMSE 2025 Demystifying LLM-Based Software Engineering Agents. [paper]
ACL 2025 Table-Critic: A Multi-Agent Framework for Collaborative Criticism and Refinement in Table Reasoning. [paper] [code]
arXiv 2024 AgentCoder: Multi-Agent-based Code Generation with Iterative Testing and Optimisation. [paper]
arXiv 2024 Code Generation with AlphaCodium: From Prompt Engineering to Flow Engineering. [paper] [code]
ICLR 2024 Teaching Large Language Models to Self-Debug. [paper]
ESEC/FSE 2023 Baldur: Whole-Proof Generation and Repair with Large Language Models. [paper]
NeurIPS 2023 Describe, Explain, Plan and Select: Interactive Planning with LLMs Enables Open-World Multi-Task Agents. [paper]
CoRL 2023 Inner Monologue: Embodied Reasoning through Planning with Language Models. [paper] [project]
NeurIPS 2023 Self-Refine: Iterative Refinement with Self-Feedback. [paper]

Search, Verification, and Acceptance

arXiv 2026 CoSPlay: Cooperative Self-Play at Test-Time with Self-Generated Code and Unit Test. [paper] [code]
ACL 2026 Findings Inference-Time Scaling of Verification: Self-Evolving Deep Research Agents via Test-Time Rubric-Guided Verification (DeepVerifier). [paper] [code]
Nature Communications 2026 Reasoning in Machine Vision by Learning Fast and Slow Thinking. [paper]
ICML 2026 Reasoning on the Manifold: Bidirectional Consistency for Self-Verification in Diffusion Language Models (BMC). [paper]
arXiv 2026 World-Model-Augmented Web Agents with Action Correction. [paper]
arXiv 2025 CodeMonkeys: Scaling Test-Time Compute for Software Engineering. [paper]
ICML 2025 Forest-of-Thought: Scaling Test-Time Compute for Enhancing LLM Reasoning. [paper] [code]
TMLR 2025 Is Your LLM Secretly a World Model of the Internet? Model-Based Planning for Web Agents. [paper]
ICLR 2025 Mutual Reasoning Makes Smaller LLMs Stronger Problem-Solver. [paper]
EMNLP 2025 Findings S*: Test Time Scaling for Code Generation. [paper] [code]
EMNLP 2025 START: Self-taught Reasoner with Tools. [paper]
NeurIPS 2024 Buffer of Thoughts: Thought-Augmented Reasoning with Large Language Models. [paper] [code]
ACL 2024 Findings Chain-of-Verification Reduces Hallucination in Large Language Models. [paper]
ICLR 2024 CRITIC: LLMs Can Self-Correct with Tool-Interactive Critiquing. [paper] [code]
AAAI 2024 Graph of Thoughts: Solving Elaborate Problems with Large Language Models. [paper]
ICML 2024 Language Agent Tree Search Unifies Reasoning, Acting, and Planning in Language Models. [paper] [code]
EMNLP 2024 Large Language Models Can Self-Correct with Key Condition Verification. [paper] [project]
arXiv 2024 Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters. [paper] [third-party code]
ICLR 2024 Solving Challenging Math Word Problems Using GPT-4 Code Interpreter with Code-based Self-Verification. [paper]
ICLR 2023 CodeT: Code Generation with Generated Tests. [paper]
EMNLP 2023 Findings Large Language Models are Better Reasoners with Self-Verification. [paper] [code]
ICML 2023 LEVER: Learning to Verify Language-to-Code Generation with Execution. [paper] [code]
ACL 2023 RARR: Researching and Revising What Language Models Say, Using Language Models. [paper]
EMNLP 2023 Reasoning with Language Model is Planning with World Model. [paper]
ICLR 2023 Self-Consistency Improves Chain of Thought Reasoning in Language Models. [paper] [third-party code] [project]
NeurIPS 2023 Self-Evaluation Guided Beam Search for Reasoning. [paper] [project]
NeurIPS 2023 Tree of Thoughts: Deliberate Problem Solving with Large Language Models. [paper] [code]
EMNLP 2022 Natural Language to Code Translation with Execution. [paper] [code]

Reliability and the Persistence Limit

ICLR 2024 Is Self-Repair a Silver Bullet for Code Generation? [paper]
ICLR 2024 Large Language Models Cannot Self-Correct Reasoning Yet. [paper]

🧠 L1: Model-Level Self-Evolution _{↑ contents}

Deepest active evolution target: Trainable model or policy state. Characteristic failure: Model collapse.

The three training relations of Model-Level Self-Evolution

Figure 4. The three training relations, read left to right as the party emitting the training signal moves further from the trainee and the signal becomes harder to fabricate.

↪️ Jump to: Single-Model Self-Training (51) · Competitive Self-Play (41) · Cooperative Co-Evolution (34) · Reliability and the Fixed-Scaffold Limit (11)

Single-Model Self-Training

ICML 2026 Workshop ASH: ASH: Agents that Self-Hone via Embodied Learning. [paper]
arXiv 2026 CoTEvol: COTEVOL: Self-Evolving Chain-of-Thoughts for Data Synthesis in Mathematical Reasoning. [paper]
ICML 2026 CPMobius: CPMöbius: Iterative Coach–Player Reasoning for Data-Free Reinforcement Learning. [paper] [code]
arXiv 2026 DARE: DARE: Difficulty-Adaptive Reinforcement Learning with Co-Evolved Difficulty Estimation. [paper] [code]
ACL 2026 Findings EasyRL: Easy Samples Are All You Need: Self-Evolving LLMs via Data-Efficient Reinforcement Learning. [paper] [code]
ACL 2026 EvoCoT: EvoCoT: Overcoming the Exploration Bottleneck in Reinforcement Learning for LLMs. [paper] [code]
ICLR 2026 EvoQuality: Self-Evolving Vision-Language Models for Image Quality Assessment via Voting and Ranking. [paper] [code]
arXiv 2026 EvoStreaming: EvoStreaming: Your Offline Video Model Is a Natively Streaming Assistant. [paper] [code]
arXiv 2026 Geometric Logic Consistency: Self-Evolving Spatial Reasoning in Vision Language Models via Geometric Logic Consistency. [paper]
arXiv 2026 Kairos: A Regret-Aware Native World-Action Model Stack for Physical AI. [paper] [code]
arXiv 2026 LangRetrieval: Language-Guided Self-Evolving Satellite-to-Radar Retrieval via CSI-Driven Reward. [paper]
SIGKDD 2026 LC-ERD: LC-ERD: Mining Latent Logic for Self-Evolving Reasoning via Consistency-Regulated Reward Decomposition. [paper] [code]
ICML 2026 Learning to Label: A Reinforced Self-Evolving Framework for Semi-supervised Referring Expression Segmentation. [paper]
AAAI 2026 MedS³: MedS3: Towards Medical Slow Thinking with Self-Evolved Soft Dual-sided Process Supervision. [paper] [code]
arXiv 2026 MetaClaw: Just Talk — An Agent That Meta-Learns and Evolves in the Wild. [paper] [code]
arXiv 2026 OASIF: An Efficient Obfuscation-Aware Self-Improving Framework for LLM-Based Assembly Code Instruction Following and Comprehension. [paper]
ICML 2026 One-Way Policy Optimization: One-Way Policy Optimization for Self-Evolving LLMs. [paper]
arXiv 2026 PolicyLong: Towards On-Policy Context Extension. [paper]
arXiv 2026 Rethinking Continual Experience Internalization for Self-Evolving LLM Agents. [paper] [code]
arXiv 2026 RetroAgent: From Solving to Evolving via Retrospective Dual Intrinsic Feedback. [paper] [code]
arXiv 2026 Rubric-based Self-play: Bootstrapping Post-training Signals for Open-ended Tasks via Rubric-based Self-play on Pre-training Text. [paper] [code]
arXiv 2026 SearchGym: Bootstrapping Real-World Search Agents via Cost-Effective and High-Fidelity Environment Simulation. [paper] [code]
arXiv 2026 Seirênes: Seirênes: Adversarial Self-Play with Evolving Distractions for LLM Reasoning. [paper] [code]
arXiv 2026 Self-Improving 4D Perception via Self-Distillation (SelfEvo). [paper] [code] [project]
arXiv 2026 Sentinel-VLA: A Metacognitive VLA Model with Active Status Monitoring for Dynamic Reasoning and Error Recovery. [paper]
AAAI 2026 SERL: SERL: Self-Examining Reinforcement Learning on Open-Domain. [paper] [code]
arXiv 2026 Socratic-SWE: Socratic-SWE: Self-Evolving Coding Agents via Trace-Derived Agent Skills. [paper]
arXiv 2026 The Era of Real-World Human Interaction: RL from User Conversations. [paper]
arXiv 2026 TTVS: Boosting Self-Exploring Reinforcement Learning via Test-time Variational Synthesis. [paper]
arXiv 2026 UI-Mem: Self-Evolving Experience Memory for Online Reinforcement Learning in Mobile GUI Agents. [paper] [project]
ICLR 2026 (Oral) VC-STaR: Through the Lens of Contrast: Self-Improving Visual Reasoning in VLMs. [paper] [code]
ECCV 2026 VISE: Paying More Attention to Visual Tokens in Self-Evolving Large Multimodal Models. [paper] [code] [project]
arXiv 2026 World Knowledge Exploration: Training LLM Agents for Spontaneous, Reward-Free Self-Evolution via World Knowledge Exploration. [paper] [code]
ICLR 2025 Aligning Language Models with Demonstrated Feedback. [paper] [code]
ICML 2025 Diving into Self-Evolving Training: Diving into Self-Evolving Training for Multimodal Reasoning. [paper] [code] [project]
ICLR 2025 LongPO: LongPO: Long Context Self-Evolution of Large Language Models through Short-to-Long Preference Optimization. [paper] [code]
EMNLP 2025 Middo: Model-Informed Dynamic Data Optimization for Enhanced LLM Fine-Tuning via Closed-Loop Learning. [paper] [code]
NeurIPS 2025 MindGYM: MindGYM: What Matters in Question Synthesis for Thinking-Centric Fine-Tuning? [paper] [code]
arXiv 2025 Process-based Self-Rewarding Language Models. [paper] [code]
NeurIPS 2025 Retrospective In-Context Learning for Temporal Credit Assignment with Large Language Models (RICOL). [paper]
NeurIPS 2025 Workshop RoiRL: RoiRL: Efficient, Self-Supervised Reasoning with Offline Iterative Reinforcement Learning. [paper]
CoLM 2025 SCRIT: Self-Evolving Critique Abilities in Large Language Models. [paper]
ICLR 2025 SER: Self-Evolved Reward Learning for LLMs. [paper] [project]
arXiv 2025 TTRL: Test-Time Reinforcement Learning. [paper] [code]
ICML 2024 RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedback. [paper] [third-party code]
arXiv 2024 Self-Rewarding Language Models. [paper] [third-party code]
EMNLP 2023 Large Language Models Can Self-Improve. [paper]
arXiv 2023 Reinforced Self-Training (ReST) for Language Modeling. [paper] [third-party code]
ACL 2023 Self-Instruct: Aligning Language Models with Self-Generated Instructions. [paper] [code]
arXiv 2022 Constitutional AI: Harmlessness from AI Feedback. [paper] [code] [project]
NeurIPS 2022 STaR: Bootstrapping Reasoning With Reasoning (Self-Taught Reasoner). [paper] [code]

Competitive Self-Play

ICLR 2026 Workshop ACE (Coding): ACE: Self-Evolving LLM Coding Framework via Adversarial Unit Test Generation and Preference Optimization. [paper]
arXiv 2026 Ask-Solve-Generate: Self-Evolving Unified Multimodal Understanding and Generation via Self-Consistency Rewards. [paper] [code] [project]
ECCV 2026 C2-Evo（SyncLoop）: SyncLoop: A Multimodal Dual-Loop Framework for Self-Improving Mathematical Reasoning. [paper] [code]
ACL 2026 CoEvolve: CoEvolve: Training LLM Agents via Agent-Data Mutual Evolution. [paper]
ACL 2026 DEPT: Breaking the Impasse: Dual-Scale Evolutionary Policy Training for Social Language Agents. [paper]
arXiv 2026 EvoVid: EvoVid: Temporal-Centric Self-Evolution for Video Large Language Models. [paper] [project]
ACL 2026 FoPO: Foresight Optimization for Strategic Reasoning in Large Language Models. [paper] [code]
arXiv 2026 G-Zero: G-Zero: Self-Play for Open-Ended Generation from Zero Data. [paper] [code]
ICLR 2026 Workshop (Spotlight) GASP: GASP: Guided Asymmetric Self-Play for Coding LLMs. [paper]
ICLR 2026 Generative Adversarial Reasoner: Generative Adversarial Reasoner: Enhancing LLM Reasoning with Adversarial Reinforcement Learning. [paper]
arXiv 2026 GeoX: GeoX: Mastering Geospatial Reasoning Through Self-Play and Verifiable Rewards. [paper]
ACL 2026 Findings iReasoner: IREASONER: Trajectory-Aware Intrinsic Reasoning Supervision for Self-Evolving Large Multimodal Models. [paper] [code] [project]
arXiv 2026 IRIS: Interpolative Rényi Iterative Self-play for Large Language Model Fine-Tuning. [paper]
arXiv 2026 KG Paths as Supervision: Knowledge-Graph Paths as Intermediate Supervision for Self-Evolving Search Agents. [paper]
ICLR 2026 R-Zero: Self-Evolving Reasoning LLM from Zero Data. [paper]
arXiv 2026 RISE: RISE: Reliable Improvement in Self-Evolving Vision-Language Models. [paper] [code]
ICML 2026 S-SPPO: S-SPPO: Semantic-Calibrated Self-Play Preference Optimization. [paper] [code]
arXiv 2026 SEIF: SEIF: Self-Evolving Reinforcement Learning for Instruction Following. [paper] [code]
arXiv 2026 SELF-EMO: Emotional Self-Evolution from Recognition to Consistent Expression. [paper]
arXiv 2026 Self-Evolving Visual Questioner. [paper] [code] [project]
ICML 2026 Self-Play Only Evolves When Self-Synthetic Pipeline Ensures Learnable Information Gain. [paper]
ICML 2026 Self-play SWE-RL: Toward Training Superintelligent Software Agents through Self-Play SWE-RL. [paper]
arXiv 2026 SpatialEvo: SpatialEvo: Self-Evolving Spatial Intelligence via Deterministic Geometric Environments. [paper] [code]
ICLR 2026 SPELL: SPELL: Self-Play Reinforcement Learning for Evolving Long-Context Language Models. [paper] [code]
ICLR 2026 SPIRAL: SPIRAL: Self-Play on Zero-Sum Games Incentivizes Reasoning via Multi-Agent Multi-Turn Reinforcement Learning. [paper] [code]
arXiv 2026 Survive or Collapse: Survive or Collapse: The Asymmetric Roles of Data Gating and Reward Grounding in Self-Play RL. [paper]
ACL 2026 TPAW: Team-Based Self-Play With Dual Adaptive Weighting for Fine-Tuning LLMs. [paper] [code]
ICML 2026 Transitivity Meets Cyclicity: Explicit Preference Decomposition for Dynamic Large Language Model Alignment. [paper] [code]
ICML 2026 TSP（Tree-like Self-Play）: Learn from Your Mistakes: Tree-like Self-Play for Secure Code LLMs. [paper] [code]
ICLR 2026 Vision-Zero: Vision-Zero: Scalable VLM Self-Evolution via Multi-Agent Self-Play. [paper] [code]
CVPR 2026 VisPlay: Self-Evolving Vision-Language Models. [paper]
arXiv 2026 Vocabulary Dropout for Curriculum Diversity in LLM Co-Evolution. [paper] [project]
arXiv 2025 Absolute Zero: Reinforced Self-play Reasoning with Zero Data. [paper] [code] [project]
ICML 2025 CDG: Improving Rationality in the Reasoning Process of Language Models through Self-playing Game. [paper]
ICLR 2025 Magnetic Preference Optimization: Achieving Last-Iterate Convergence for Language Model Alignment. [paper]
NeurIPS 2025 SPACE: SPACE: Noise Contrastive Estimation Stabilizes Self-Play Fine-Tuning for Large Language Models. [paper]
EMNLP 2025 Findings SPFT-SQL: SPFT-SQL: Enhancing Large Language Model for Text-to-SQL Parsing by Self-Play Fine-Tuning. [paper]
NeurIPS 2025 T-SPIN: Triplets Better Than Pairs: Towards Stable and Effective Self-Play Fine-Tuning for LLMs. [paper]
ACL 2025 Findings TRANS-ZERO: TRANS-ZERO: Self-Play Incentivizes Large Language Models for Multilingual Translation Without Parallel Data. [paper] [code]
NeurIPS 2024 Adversarial Taboo Self-Play (SPAG): Self-playing Adversarial Language Game Enhances LLM Reasoning. [paper] [code]
ICML 2024 SPIN: Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models. [paper] [code] [project]

Cooperative Co-Evolution

arXiv 2026 ARISE: Agent Reasoning with Intrinsic Skill Evolution in Hierarchical Reinforcement Learning. [paper] [code]
arXiv 2026 Be My Tutor: On-Policy Co-Distillation for Mutual LLM Improvement via Peer Feedback. [paper]
arXiv 2026 Co-Evolution of Policy and Internal Reward for Language Agents. [paper]
arXiv 2026 Co-Evolving Policy Distillation: Co-Evolving Policy Distillation. [paper]
arXiv 2026 CoHyDE: Iterative Co-Training of LLM Rewriter & Dense Encoder for Tool Retrieval. [paper]
arXiv 2026 COMAP: COMAP: Co-Evolving World Models and Agent Policies for LLM Agents. [paper] [code]
arXiv 2026 Cross-Model Entropy: Label-Free Reinforcement Learning via Cross-Model Entropy. [paper]
arXiv 2026 DUEL: Adversarial Self-Play for Multimodal Reasoning. [paper]
arXiv 2026 Evolving-RL: Evolving-RL: End-to-End Optimization of Experience-Driven Self-Evolving Capability within Agents. [paper] [code]
ICML 2026 From blind spots to gains: Diagnostic-driven iterative training for large multimodal models. [paper] [code]
arXiv 2026 GenEvolve: Self-Evolving Image Generation Agents via Tool-Orchestrated Visual Experience Distillation. [paper] [code] [project]
arXiv 2026 In-the-Flow Agentic System Optimization for Effective Planning and Tool Use (AgentFlow). [paper] [code] [project]
arXiv 2026 Interactive Critique-Revision Training for Reliable Structured LLM Generation. [paper]
LREC 2026 Workshop Learning to Negotiate: Multi-Agent Deliberation for Collective Value Alignment in LLMs. [paper]
ACL 2026 MAESTRO: Meta-learning Adaptive Estimation of Scalarization Trade-offs for Reward Optimization. [paper] [code]
arXiv 2026 Measure Twice, Click Once: Measure Twice, Click Once: Co-evolving Proposer and Visual Critic via Reinforcement Learning for GUI Grounding. [paper]
arXiv 2026 OPD-Evolver: Cultivating Holistic Agent Evolver via On-Policy Distillation. [paper] [code]
arXiv 2026 PopuLoRA: PopuLoRA: Co-Evolving LLM Populations for Reasoning Self-Play. [paper]
arXiv 2026 Programming with Data: Test-Driven Data Engineering for Self-Improving LLMs from Raw Corpora. [paper] [code]
arXiv 2026 SEAgent: Self-Evolving Computer Use Agent with Autonomous Learning from Experience. [paper] [code]
arXiv 2026 Self-Distilled RL: Self-Distilled Reinforcement Learning for Co-Evolving Agentic Recommender Systems. [paper]
ICML 2026 Self-evolving LLM Agents with In-distribution Optimization (Q-Evolve). [paper] [project]
ICLR 2026 (Poster) Self-Harmony: Learning to Harmonize Self-Supervision and Self-Play in Test-Time Reinforcement Learning. [paper] [code]
ACL 2026 Findings SERM: SERM: Self-Evolving Relevance Model with Agent-Driven Learning from Massive Query Streams. [paper]
arXiv 2026 Variational Policy Distillation: Learning from Language Feedback via Variational Policy Distillation. [paper]
arXiv 2026 ZeroCoder: Can LLMs Improve Code Generation Without Ground-Truth Supervision? [paper]
arXiv 2026 π-Play: Multi-Agent Self-Play via Privileged Self-Distillation without External Data. [paper] [code]
CoLM 2025 Collaborative self-play（元知识）: Don't lie to your friends: Learning what you know from collaborative self-play. [paper]
NeurIPS 2025 (Spotlight) CURE: Co-Evolving LLM Coder and Unit Tester via Reinforcement Learning. [paper] [code]
NeurIPS 2025 QiMeng-MuPa: QiMeng-MuPa: Mutual-Supervised Learning for Sequential-to-Parallel Code Translation. [paper] [code]
NeurIPS 2025 RL Tango: RL Tango: Reinforcing Generator and Verifier Together for Language Reasoning. [paper] [code]
EMNLP 2025 WebEvolver: Enhancing Web Agent Self-Improvement with Co-evolving World Model. [paper] [code]
ICLR 2025 WebRL: Training LLM Web Agents via Self-Evolving Online Curriculum Reinforcement Learning. [paper] [code]
NeurIPS 2024 CORY: Coevolving with the Other You: Fine-Tuning LLM with Sequential Cooperative Multi-Agent Reinforcement Learning. [paper] [code]

Reliability and the Fixed-Scaffold Limit

arXiv 2026 Autonomous Drift Learning: Autonomous Drift Learning in Data Streams: A Unified Perspective. [paper]
arXiv 2026 Confidence-Orchestrated Self-Evolution: Confidence-Orchestrated Self-Evolution against Uncertain LLM Feedback. [paper] [code]
arXiv 2026 Do Self-Evolving Agents Forget?: Do Self-Evolving Agents Forget? Capability Degradation and Preservation in Lifelong LLM Agent Adaptation. [paper]
arXiv 2026 First-Order Recoverability Collapse: First-Order Recoverability Collapse in Self-Referential Information Decoders. [paper]
arXiv 2026 Implicit Conflict Monitoring: Modeling Implicit Conflict Monitoring Mechanisms Against Stereotypes in LLMs. [paper]
arXiv 2026 Matrix-Level Dynamics: When Self-Reference Fails to Close: Matrix-Level Dynamics in Large Language Models. [paper]
ICML 2026 On the Generalization Gap in Self-Evolving Language Model Reasoning. [paper]
NAACL 2025 GSI: Mitigating Tail Narrowing in LLM Self-Improvement via Socratic-Guided Sampling. [paper] [code]
NeurIPS 2025 Is PRM Necessary?: Is PRM Necessary? Problem-Solving RL Implicitly Induces PRM Capability in LLMs. [paper]
EMNLP 2025 Superficial Self-Improved Reasoners: Superficial Self-Improved Reasoners Benefit from Model Merging. [paper]
ICML 2024 A Tale of Tails: A Tale of Tails: Model Collapse as a Change of Scaling Laws. [paper]

🧰 L2: Scaffold-Level Self-Evolution _{↑ contents}

Deepest active evolution target: Scaffold. Characteristic failure: Scaffold overfitting.

The widening scaffold scope of Scaffold-Level Self-Evolution

Figure 5. The widening scaffold scope, from a single prompt or code artifact out to the runtime harness that encloses them all. Each wider region presupposes the narrower objects it organizes, while the improver and criterion stay fixed.

↪️ Jump to: Definition and the Scaffold Boundary (4) · Prompts and Programs (17) · Architecture and Workflows (77) · Skills and Experience (106) · Memory and Retrieval (31) · Runtime Harness (14) · Reliability and the Fixed-Improver Limit (8)

Definition and the Scaffold Boundary

arXiv 2026 Darwin Mobile Agent: Darwin Mobile Agent: A Roadmap for Self-Evolution. [paper]
arXiv 2026 From Chatbot to Digital Colleague: From Chatbot to Digital Colleague: The Paradigm Shift Toward Persistent Autonomous AI. [paper] [project]
arXiv 2026 Next-Gen Agentic RL Systems: Next-Generation Agentic Reinforcement Learning Systems Enable Self-Evolving Agents. [paper] [code]
arXiv 2026 Root Theorem of Context Engineering: The Root Theorem of Context Engineering: Formal Derivation, Architectural Prediction, and Engineering Proof. [paper]

Prompts and Programs

ICASSP 2026 AutoVQA-G: AutoVQA-G: Self-Improving Agentic Framework for Automated Visual Question Answering and Grounding Annotation. [paper] [code]
arXiv 2026 Bi-Component AHD: LLM-Driven Co-Evolutionary Automated Heuristic Design for Bi-Component Coupled Combinatorial Optimization. [paper]
arXiv 2026 Combee: Combee: Scaling Prompt Learning for Self-Improving Language Model Agents. [paper] [code]
arXiv 2026 DataEvolver: Automatic Data Preparation for Large Language Models through Multi-Level Self-Evolving. [paper] [code]
ICML 2026 Workshop Dense Feedback for Social Dilemmas: Beyond Scalar Rewards: Dense Feedback for LLM Policy Synthesis in Sequential Social Dilemmas. [paper] [code]
arXiv 2026 EEVEE: Eevee: Towards Test-time Prompt Learning in the Real World for Self-Improving Agents. [paper] [code] [project]
arXiv 2026 Fluid Control Discovery: Self-Evolving Scientific Agent Discovers Generalizable Physically-Reasoned Fluid Control. [paper]
arXiv 2026 GenTI: Benchmarking LLMs for Autonomous IDPS Rule Generation for Unseen Attacks. [paper]
CVPR 2026 HIER: Evolutionary Multimodal Reasoning via Hierarchical Semantic Representation for Intent Recognition. [paper] [code]
arXiv 2026 InferenceEvolve: InferenceEvolve: Automated Causal Effect Estimators through Self-Evolving AI. [paper] [code] [project]
arXiv 2026 MCE (Meta Context Engineering): Meta Context Engineering via Agentic Skill Evolution. [paper] [code]
arXiv 2026 MLEvolve: MLEvolve: A Self-Evolving Framework for Automated Machine Learning Algorithm Discovery. [paper] [code]
DAC 2026 Multi-Agent Self-Evolved ABC: Autonomous Evolution of EDA Tools: Multi-Agent Self-Evolved ABC. [paper]
ECIR Workshop 2026 Workshop Self-Optimizing MAS for Deep Research: Self-Optimizing Multi-Agent Systems for Deep Research. [paper]
SEAMS 2026 SelfEvolve: Software Self-Extension with SelfEvolve: an Agentic Architecture for Runtime Code Generation. [paper]
arXiv 2026 SHARP: SHARP: A Self-Evolving Human-Auditable Rubric Policy for Financial Trading Agents. [paper]
ICML 2025 Workshop Game-Playing via Generative Code Optimization: Learning Game-Playing Agents with Generative Code Optimization. [paper] [code]

Architecture and Workflows

arXiv 2026 A Self-Evolving Agentic Framework for Metasurface Inverse Design. [paper]
arXiv 2026 A Self-Evolving Agentic System for Automated Generation and Execution of Biological Protocols (ProtoPilot). [paper]
arXiv 2026 ABot-Claw: A Foundation for Persistent, Cooperative, and Self-Evolving Robotic Agents. [paper]
arXiv 2026 Agent libOS: A Runtime Substrate for Capability-Controlled Self-Evolving LLM Agents. [paper] [code]
arXiv 2026 AgentFactory: A Self-Evolving Framework Through Executable Subagent Accumulation and Reuse. [paper] [code]
arXiv 2026 Agentic Hardware Design as Repository-Level Code Evolution (HORIZON). [paper]
arXiv 2026 Agentic Harness Engineering: Observability-Driven Automatic Evolution of Coding-Agent Harnesses. [paper] [code]
arXiv 2026 AgentX: Towards Agent-Driven Self-Iteration of Industrial Recommender Systems. [paper]
arXiv 2026 AgRefactor: Self-Evolving Agentic Workflow for HLS Compatibility and Performance. [paper]
arXiv 2026 AIRA-Compose / AIRA-Design: Agentic Discovery of Neural Architectures: AIRA-Compose and AIRA-Design. [paper]
arXiv 2026 ArtiCAD: Articulated CAD Assembly Design via Multi-Agent Code Generation. [paper]
arXiv 2026 Autogenesis: A Self-Evolving Agent Protocol. [paper] [code]
arXiv 2026 Autopoiesis: A Self-Evolving System Paradigm for LLM Serving Under Runtime Dynamics. [paper]
arXiv 2026 Bian Que: An Agentic Framework with Flexible Skill Arrangement for Online System Operations. [paper] [code] [project]
arXiv 2026 BloClaw: An Omniscient, Multi-Modal Agentic Workspace for Next-Generation Scientific Discovery. [paper] [code]
arXiv 2026 Catalyst Discovery: Autonomous Heterogeneous Catalyst Discovery with a Self-Evolving Multi-Agent Digital Twin. [paper]
arXiv 2026 Co-evolving Agent Architectures and Interpretable Reasoning for Automated Optimization. [paper]
arXiv 2026 Compute Allocation in Evo Search: Compute Allocation in Evolutionary Search: From Depth–Breadth to Multi-Armed Bandits. [paper] [code]
arXiv 2026 CyberEvolver: Structured Self-Evolution for Cybersecurity Agents On the Fly. [paper]
arXiv 2026 Differentiable Mixture-of-Agents Incentivizes Swarm Intelligence of Large Language Models. [paper]
arXiv 2026 EGL-SCA: Structural Credit Assignment for Co-Evolving Instructions and Tools in Graph Reasoning Agents. [paper]
arXiv 2026 EpiEvolve: Self-Evolving Agents for Streaming Pandemic Forecasting under Regime Shifts. [paper]
IEEE TCAD 2026 Evidence-Driven LLM Agent for C-to-Synthesizable-C Conversion and Verification. [paper]
arXiv 2026 EVOCHAMBER: Test-Time Co-evolution of Multi-Agent System at Individual, Team, and Population Scales. [paper] [code]
arXiv 2026 EvoDrive: Pareto Evolution for Safety-Critical Autonomous Driving via Self-Improving LLM Agents. [paper] [project]
arXiv 2026 EvolveRouter: Co-Evolving Routing and Prompt for Multi-Agent Question Answering. [paper]
arXiv 2026 EvoMaster: A Foundational Evolving Agent Framework for Agentic Science at Scale. [paper] [code]
arXiv 2026 EvoRAG: Making Knowledge Graph-based RAG Automatically Evolve through Feedback-driven Backpropagation. [paper] [code]
ICLR 2026 EvoTest: Evolutionary Test-Time Learning for Self-Improving Agentic Systems. [paper]
arXiv 2026 Experience as a Compass: Multi-agent RAG with Evolving Orchestration and Agent Prompts. [paper]
arXiv 2026 Expert Knowledge + Feature Eng: Bridging Expert Knowledge and Automated Feature Engineering via Self-Evolution. [paper]
arXiv 2026 GRAFT-ATHENA: Self-Improving Agentic Teams for Autonomous Discovery and Evolutionary Numerical Algorithms. [paper]
arXiv 2026 GraphMind: From Operational Traces to Self-Evolving Workflow Automation. [paper]
arXiv 2026 Group-Evolving Agents: Open-Ended Self-Improvement via Experience Sharing. [paper] [code]
arXiv 2026 Learning to Evolve: A Self-Improving Framework for Multi-Agent Systems via Textual Parameter Graph Optimization. [paper]
arXiv 2026 MetaGen: Self-Evolving Roles and Topologies for Multi-Agent LLM Reasoning. [paper]
arXiv 2026 Mimosa Framework: Toward Evolving Multi-Agent Systems for Scientific Research. [paper]
arXiv 2026 MobEvolve: An Agentic Self-Evolving Heuristic System for Interpretable Human Mobility Generation. [paper]
arXiv 2026 MUSE: Multi-Domain Chinese User Simulation via Self-Evolving Profiles and Rubric-Guided Alignment. [paper]
AAAI 2026 NOTAM-Evolve: NOTAM-Evolve: A Knowledge-Guided Self-Evolving Optimization Framework with LLMs for NOTAM Interpretation. [paper]
arXiv 2026 OctoT2I: A Self-Evolving Agentic Text-to-Image Router. [paper]
arXiv 2026 PACE: Two-Timescale Self-Evolution for Small Language Model Agents. [paper]
arXiv 2026 Parthenon Law: A Self-Evolving Legal-Agent Framework. [paper]
ICML 2026 PathWise: PathWise: Planning through World Model for Automated Heuristic Design via Self-Evolving LLMs. [paper] [code]
arXiv 2026 PFAgent: A Tractable and Self-Evolving Power-Flow Agent for Interactive Grid Analysis. [paper]
arXiv 2026 PulseCX: Breaking the Closed-World Assumption in Real-Time CX. [paper]
arXiv 2026 QueenBee Planner: Skill-Evolving Communication Topologies for Token-Efficient LLM Multi-Agent Systems. [paper]
arXiv 2026 RewardHarness: Self-Evolving Agentic Post-Training. [paper] [code] [project]
arXiv 2026 RFAmpDesigner: A Self-Evolving Multi-Agent LLM Framework for Automated Radio Frequency Amplifier Design. [paper]
arXiv 2026 Roles with Rails: Contract-Preserving Role Evolution in Multi-Agent Structured Reasoning. [paper]
ACL 2026 SEARL: SEARL: Joint Optimization of Policy and Tool Graph Memory for Self-Evolving Agents. [paper] [code]
arXiv 2026 Self-Evolving Agentic Image Restoration via Deliberate Planning and Intuitive Execution (SEAR). [paper]
AAMAS 2026 Self-Evolving Software Agents: Self-Evolving Software Agents (Extended Abstract). [paper]
ACL 2026 Findings SEMA-RAG: SEMA-RAG: A Self-Evolving Multi-Agent Retrieval-Augmented Generation Framework for Medical Reasoning. [paper]
arXiv 2026 SkillGraph: Self-Evolving Multi-Agent Collaboration with Multimodal Graph Topology. [paper] [code]
arXiv 2026 SpaceMind: A Modular and Self-Evolving Embodied Vision-Language Agent Framework for Autonomous On-orbit Servicing. [paper] [code]
arXiv 2026 TabClaw: An Interactive and Self-Evolving Agent for Spreadsheet Manipulation and Table Reasoning. [paper] [code] [project]
arXiv 2026 TacEvo: TacEvo: Self-Evolving Architecture Discovery for Robotic Tactile Perception via LLM-Driven Quality-Diversity Search. [paper]
arXiv 2026 TacoMAS: Test-Time Co-Evolution of Topology and Capability in LLM-based Multi-Agent Systems. [paper] [code]
arXiv 2026 The Log is the Agent: Event-Sourced Reactive Graphs for Auditable, Forkable Agentic Systems. [paper]
arXiv 2026 TopoEvo: A Topology-Aware Self-Evolving Multi-Agent Framework for Root Cause Analysis in Microservices. [paper]
IEEE Communications Magazine 2026 Toward Intelligent and Secure Cloud: Large Language Model Empowered Proactive Defense (LLM-PD). [paper] [code]
arXiv 2026 Toward Vibe Medicine: A Self-Evolving Multi-Agent Framework for Clinical Decision Support. [paper]
arXiv 2026 Towards Recursive Self-Evolving Agentic Literature Retrieval. [paper] [code]
arXiv 2026 Traj-Evolve: A Self-Evolving Multi-Agent System for Patient Trajectory Modeling in Lung Cancer Early Detection. [paper]
ICLR 2026 Workshop Universe Routing: Universe Routing: Why Self-Evolving Agents Need Epistemic Control. [paper]
arXiv 2026 VisualClaw: A Real-Time, Personalized Agent for the Physical World. [paper] [code] [project]
arXiv 2026 Web2BigTable: A Bi-Level Multi-Agent LLM System for Internet-Scale Information Search and Extraction. [paper] [code]
arXiv 2025 AgentOrchestra: Orchestrating multi-agent intelligence with the tool-environment-agent(TEA) protocol. [paper]
ICLR 2025 AgentSquare: Automatic LLM Agent Search in Modular Design Space. [paper] [code] [project]
NeurIPS 2025 C-NAV: Towards Self-Evolving Continual Object Navigation in Open World. [paper] [code] [project]
NeurIPS 2025 (Oral) MAS-ZERO: Designing Multi-Agent Systems with Zero Supervision. [paper] [code]
MICCAI 2025 (Oral) MedAgentSim: Self-Evolving Multi-Agent Simulations for Realistic Clinical Interactions. [paper] [code] [project]
KDD 2025 MobileSteward: Integrating Multiple App-Oriented Agents with Self-Evolution to Automate Cross-App Instructions. [paper]
NeurIPS 2025 Multi-Agent Collaboration via Evolving Orchestration. [paper] [code]
SIGMOD 2025 SEFRQO: A Self-Evolving Fine-Tuned RAG-Based Query Optimizer. [paper]
ICML 2024 GPTSwarm: Language Agents as Optimizable Graphs. [paper]

Skills and Experience

arXiv 2026 A Self-Evolving Framework for Efficient Terminal Agents via Observational Context Compression. [paper] [code]
arXiv 2026 Ace-Skill: ACE-SKILL: Bootstrapping Multimodal Agents with Prioritized and Clustered Evolution. [paper] [code]
arXiv 2026 AgenticRecTune: Multi-Agent with Self-Evolving Skillhub for Recommendation System Optimization. [paper]
arXiv 2026 AlphaMemo: Structured Search-Process Memory for Self-Evolving Alpha Mining Agents. [paper] [code]
arXiv 2026 ANNEAL: ANNEAL: Adapting LLM Agents via Governed Symbolic Patch Learning. [paper] [code]
arXiv 2026 APEX: Autonomous Policy Exploration for Self-Evolving LLM Agents. [paper] [code]
arXiv 2026 AtlasVA: Self-Evolving Visual Skill Memory for Teacher-Free VLM Agents. [paper] [code] [project]
arXiv 2026 AutoSkill: AutoSkill: Experience-Driven Lifelong Learning via Skill Self-Evolution. [paper] [code]
arXiv 2026 Beyond Meta-Reasoning: Metacognitive Consolidation for Self-Improving LLM Reasoning. [paper]
arXiv 2026 Causal World Modeling: Self-Evolving Cognitive Framework via Causal World Modeling for Embodied Scientific Intelligence. [paper]
arXiv 2026 Co-Evolving LLM Decision and Skill Bank Agents for Long-Horizon Tasks. [paper] [code] [project]
arXiv 2026 CoCoDA: CoCoDA: Co-evolving Compositional DAG for Tool-Augmented Agents. [paper]
arXiv 2026 CoEvoSkills: Self-Evolving Agent Skills via Co-Evolutionary Verification. [paper] [code] [project]
arXiv 2026 COMFYCLAW: COMFYCLAW: Self-Evolving Skill Harnesses for Image Generation Workflows. [paper]
arXiv 2026 Decentralized Memory: Self-Evolving Multi-Agent Systems via Decentralized Memory. [paper]
arXiv 2026 DeliCIR: Deliberative Test-Time Evolutionary Hierarchical Multi-Agents for Composed Image Retrieval. [paper]
arXiv 2026 Detect in Any Scene: An Agentic Framework for Object Detection with Experience-Aware Reasoning. [paper]
ICML 2026 DocOS: Towards Proactive Document-Guided Actions in GUI Agents. [paper]
arXiv 2026 DrugSAGE: Self-evolving Agent Experience for Efficient State-of-the-Art Drug Discovery. [paper]
arXiv 2026 Dual-Process Cognitive Memory: Memory Beyond Recall: A Dual-Process Cognitive Memory System for Self-Evolving LLM Agents. [paper]
IJCNN 2026 EEAgent: Evolvable Embodied Agent for Robotic Manipulation via Long Short-Term Reflection and Optimization. [paper]
arXiv 2026 EmbodiSkill: EmbodiSkill: Skill-Aware Reflection for Self-Evolving Embodied Agents. [paper] [code]
ICLR 2026 Workshop ERL: Experiential Reflective Learning for Self-Improving LLM Agents. [paper]
arXiv 2026 ESAA-Conversational: ESAA-Conversational: An Event-Sourced Memory Layer for Continuity, Handoff, and Curation Across Heterogeneous LLM Coding Agents. [paper]
arXiv 2026 ESC-Skills: Discovering and Self-Evolving Skills for Emotional Support Conversations. [paper] [code]
arXiv 2026 Evo-MedAgent: Beyond One-Shot Diagnosis with Agents That Remember, Reflect, and Improve. [paper]
KDD 2026 EvoDS: EvoDS: Self-Evolving Autonomous Data Science Agent with Skill Learning and Context Management. [paper] [code]
arXiv 2026 EvoIR-Agent: Self-Evolving Image Restoration Agentic System via Experience-Driven Learning. [paper]
arXiv 2026 EvolveMem: Self-Evolving Memory Architecture via AutoResearch for LLM Agents. [paper] [code]
arXiv 2026 EvolveNav: Proactive Preflection and Self-Evolving Memory for Zero-Shot Object Goal Navigation. [paper]
ICML 2026 EvolveR: EvolveR: Self-Evolving LLM Agents Through an Experience-Driven Lifecycle. [paper] [code]
arXiv 2026 EvoMemNav: EvoMemNav: Efficient Self-Evolving Fine-Grained Memory for Zero-Shot Embodied Navigation. [paper] [code]
arXiv 2026 EvoRec: Self-Evolving Agentic Recommender Systems. [paper]
arXiv 2026 EvoRepair: Enhancing Vulnerability Repair Agents Through Experience-Based Self-Evolution. [paper]
arXiv 2026 EvoSkill: EvoSkill: Automated Skill Discovery for Multi-Agent Systems. [paper] [code]
arXiv 2026 EXG: Self-Evolving Agents with Experience Graphs. [paper]
arXiv 2026 Experience Graphs: Experience Graphs: The Data Foundation for Self-Improving Agents. [paper]
arXiv 2026 ExpGraph: Model-Agnostic Experience Learning with Graph-Structured Memory for LLM Agents. [paper]
arXiv 2026 FederatedSkill: FederatedSkill: Federated Learning for Agentic Skill Evolution. [paper] [code]
arXiv 2026 Few-Shot MTS: Empowering VLMs for Few-Shot Multimodal Time Series Classification via Tailored Agentic Reasoning. [paper]
arXiv 2026 FinAcumen: Financial Multimodal Reasoning via Self-Evolving Experience Memory Harness. [paper] [code]
arXiv 2026 FlyRoute: Self-Evolving Agent Profiling via Data Flywheel for Adaptive Task Routing. [paper]
arXiv 2026 Forage V2: Forage V2: Knowledge Evolution and Transfer in Autonomous Agent Organizations. [paper]
arXiv 2026 FORGE: Self-Evolving Agent Memory With No Weight Updates via Population Broadcast. [paper] [code]
arXiv 2026 From Context to Skills: Can Language Models Learn from Context Skillfully? [paper] [code]
arXiv 2026 GenericAgent: A Token-Efficient Self-Evolving LLM Agent via Contextual Information Density Maximization (V1.0). [paper] [code]
arXiv 2026 GRASP: Gated Regression-Aware Skill Proposer for Self-Improving LLM Agents. [paper] [code] [project]
arXiv 2026 Health Community Notes: Better with Experience: Self-Evolving LLM Agents for Evidence-Grounded Health Community Notes. [paper]
ICML 2026 Workshop ISM: Self-Improving Strategy Memory for Continual Mathematical Reasoning. [paper] [code]
arXiv 2026 KBSpec: KBSpec: LLM-driven Formal Specification Generation with Evolving Domain Knowledge Base. [paper]
arXiv 2026 M2Note: M2Note: Continual Evolution of Vision Language Models via Mistake Notebook Learning. [paper]
arXiv 2026 MAGE: MAGE: Multi-Agent Self-Evolution with Co-Evolutionary Knowledge Graphs. [paper]
arXiv 2026 ManimAgent: Self-Evolving Multimodal Agents for Visual Education. [paper] [code] [project]
arXiv 2026 Medical Skill Memory: Experience Makes Skillful: Enabling Generalizable Medical Agent Reasoning via Self-Evolving Skill Memory. [paper]
arXiv 2026 MemQ: MemQ: Integrating Q-Learning into Self-Evolving Memory Agents over Provenance DAGs. [paper] [code]
arXiv 2026 MetaForge: MetaForge: A Self-Evolving Multimodal Agent that Retrieves, Adapts, and Forges Tools On Demand. [paper]
arXiv 2026 Metis (Text+Code Memory): METIS: Bridging Text and Code Memory for Self-Evolving Agents. [paper]
arXiv 2026 MMG2Skill: Can Agents Distill In-the-Wild Guides into Self-Evolving Skills? [paper] [code]
arXiv 2026 MUSE-Autoskill: Self-Evolving Agents via Skill Creation, Memory, Management, and Evaluation. [paper]
ICML 2026 NSI: Lifting Traces to Logic: Programmatic Skill Induction with Neuro-Symbolic Learning for Long-Horizon Agentic Tasks. [paper]
arXiv 2026 OpenSkill: OpenSkill: Open-World Self-Evolution for LLM Agents. [paper] [code]
arXiv 2026 PhotoCraft: PhotoCraft: Agentic Reasoning with Hierarchical Self-Evolving Memory for Deep Image Search. [paper]
arXiv 2026 Prism: An Evolutionary Memory Substrate for Multi-Agent Open-Ended Discovery. [paper]
arXiv 2026 ProPlay: Procedural World Models for Self-Evolving LLM Agents. [paper] [code]
arXiv 2026 PsychAgent: An Experience-Driven Lifelong Learning Agent for Self-Evolving Psychological Counselor. [paper] [code]
ICML 2026 REFLEX: Reflective Evolution from LLM Experience. [paper]
arXiv 2026 Rethinking Experience Utilization: Rethinking Experience Utilization in Self-Evolving Language Model Agents. [paper]
arXiv 2026 Retrieval as Reasoning: Self-Evolving Agent-Native Retrieval via LLM-Wiki. [paper]
arXiv 2026 Robo-Cortex: A Self-Evolving Embodied Agent via Dual-Grain Cognitive Memory and Autonomous Knowledge Induction. [paper]
arXiv 2026 SAGE (Graph-Memory): SAGE: A Self-Evolving Agentic Graph-Memory Engine for Structure-Aware Associative Memory. [paper]
arXiv 2026 SAGER: Self-Evolving User Policy Skills for Recommendation Agent. [paper]
arXiv 2026 Security Auditing Playbooks: Transferable Self-Evolving Playbooks for Agentic Security Auditing. [paper]
arXiv 2026 Self-Evolving LLM Memory Extraction Across Heterogeneous Tasks. [paper] [code]
ACL 2026 Self-Evolving Rule-Driven Legal Retrieval: When Rules Learn: A Self-Evolving Agent for Legal Case Retrieval. [paper]
arXiv 2026 Self-Evolving World Models: Self-Evolving World Models for LLM Agent Planning. [paper]
arXiv 2026 SetupX: Can LLM Agents Learn from Past Failures in Functionality-Correct Code Repository Setup? [paper] [code]
arXiv 2026 Skill-3D: Skill-3D: Evolving Scene-Aware Skills for Agentic 3D Spatial Reasoning. [paper] [code] [project]
arXiv 2026 SkillClaw: SkillClaw: Let Skills Evolve Collectively with Agentic Evolver. [paper] [code]
arXiv 2026 SkillComposer: SkillComposer: Learning to Evolve Agent Skills for Specification and Generalization. [paper]
arXiv 2026 SkillDAG: SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale. [paper]
SIGIR 2026 SkillForge: SkillForge: Forging Domain-Specific, Self-Evolving Agent Skills in Cloud Technical Support. [paper]
arXiv 2026 SKILLFOUNDRY: Building Self-Evolving Agent Skill Libraries from Heterogeneous Scientific Resources. [paper] [code] [project]
arXiv 2026 SkillOpt: Executive Strategy for Self-Evolving Agent Skills. [paper] [code] [project]
arXiv 2026 SkillOS: SkillOS: Learning Skill Curation for Self-Evolving Agents. [paper]
arXiv 2026 SkillRevise: Improving LLM-Authored Agent Skills via Trace-Conditioned Skill Revision. [paper]
arXiv 2026 Skills-Coach: A Self-Evolving Skill Optimizer via Training-Free GRPO. [paper] [code]
arXiv 2026 SkillSmith: Co-Evolving Skills and Tools for Self-Improving Agent Systems. [paper]
arXiv 2026 SkillsVote: SkillsVote: Lifecycle Governance of Agent Skills from Collection, Recommendation to Evolution. [paper] [code]
arXiv 2026 SkillWiki: A Living Knowledge Infrastructure for Agent Skills. [paper] [code]
arXiv 2026 SkillX: Automatically Constructing Skill Knowledge Bases for Agents. [paper] [code]
arXiv 2026 Swarm Skills: Swarm Skills: A Portable, Self-Evolving Multi-Agent System Specification for Coordination Engineering. [paper]
arXiv 2026 Trace2Policy: From Expert Behavior Traces to Self-Evolving Decision Agents. [paper]
arXiv 2026 Tree-of-Experience: Tree-of-Experience: A Structured Experience-Management Solution for Self-Evolving Agents under Low-Repetition and Implicit-Reward Environments. [paper]
ICRA 2026 Uni-Skill: Uni-Skill: Building Self-Evolving Skill Repository for Generalizable Robotic Manipulation. [paper]
arXiv 2026 VASO: VASO: Formally Verifiable Self-Evolving Skills for Physical AI Agents. [paper] [project]
arXiv 2026 When Search Becomes Memory: Turning Robot Design Trials into Transferable Skills. [paper]
SOSE 2025 Advanced tool learning and selection system (ATLASS): a closed-loop framework using LLM. [paper]
arXiv 2025 Alita: Generalist agent enabling scalable agentic reasoning with minimal predefinition and maximal self-evolution. [paper] [code]
ACL 2025 Contextual Experience Replay (CER): Contextual Experience Replay for Self-Improvement of Language Agents. [paper]
arXiv 2025 ELL: Building Self-Evolving Agents via Experience-Driven Lifelong Learning: A Framework and Benchmark. [paper] [code] [project]
CVPR 2025 ESCHER: Self-Evolving Visual Concept Library using Vision-Language Critics. [paper] [code] [project]
arXiv 2025 MetaAgent: MetaAgent: Toward Self-Evolving Agent via Tool Meta-Learning. [paper] [code]
arXiv 2025 SkillWeaver: Web agents can self-improve by discovering and honing skills. [paper]
arXiv 2025 STELLA: Self-evolving LLM agent for biomedical research. [paper]
ACL 2024 Self-Evolving GPT: Self-Evolving GPT: A Lifelong Autonomous Experiential Learner. [paper] [code]
TMLR 2024 Voyager: An Open-Ended Embodied Agent with Large Language Models. [paper] [code] [project]

Memory and Retrieval

ICLR 2026 ACE（Agentic Context Engineering）: Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models. [paper] [code]
ACL 2026 APEX-MEM: Agentic Semi-Structured Memory with Temporal Reasoning for Long-Term Conversational AI. [paper]
arXiv 2026 AutoAgent: AutoAgent: Evolving Cognition and Elastic Memory Orchestration for Adaptive Agents. [paper]
MICCAI 2026 Evo-RAD: Evo-RAD: Navigating Rare Retinal Disease Diagnosis via Self-Evolving Agentic Retrieval. [paper] [code]
arXiv 2026 Faulty Memories: Useful Memories Become Faulty When Continuously Updated by LLMs. [paper]
arXiv 2026 H-Mem: H-MEM: A Novel Memory Mechanism for Evolving and Retrieving Agent Memory via a Hybrid Structure. [paper]
arXiv 2026 Live-Evo: Online Evolution of Agentic Memory from Continuous Feedback. [paper] [project]
WWW 2026 MemoTime: Memory-Augmented Temporal Knowledge Graph Enhanced Large Language Model Reasoning. [paper] [code] [project]
FSE 2026 MEMRES: MemRes: A Memory-Augmented Resolver with Confidence Cascade for Agentic Python Dependency Resolution. [paper]
arXiv 2026 MemRL: Self-Evolving Agents via Runtime Reinforcement Learning on Episodic Memory. [paper] [code]
arXiv 2026 MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents. [paper] [code]
ACL 2026 Mem²Evolve: Mem2Evolve: Towards Self-Evolving Agents via Co-Evolutionary Capability Expansion and Experience Distillation. [paper] [project]
arXiv 2026 MetaMem: Evolving Meta-Memory for Knowledge Utilization through Self-Reflective Symbolic Optimization. [paper] [code]
arXiv 2026 PRIME: PRIME: Training Free Proactive Reasoning via Iterative Memory Evolution for User-Centric Agent. [paper]
ICLR 2026 ReasoningBank: Scaling Agent Self-Evolving with Reasoning Memory. [paper] [code]
ACL 2026 Findings ReMe: Remember Me, Refine Me: A Dynamic Procedural Memory Framework for Experience-Driven Agent Evolution. [paper] [code]
ICML 2026 RGMem: RGMem: Renormalization Group–inspired Memory Evolution for Language Agents. [paper] [code]
ICML 2026 SE-GA: SE-GA: Memory-Augmented Self-Evolution for GUI Agents. [paper] [code]
TMLR 2026 Thought-Retriever: Thought-Retriever: Don't Just Retrieve Raw Data, Retrieve Thoughts for Memory-Augmented Agentic Systems. [paper] [code] [project]
arXiv 2026 WebCoach: Self-Evolving Web Agents with Cross-Session Memory Guidance. [paper] [code]
arXiv 2026 xMemory: Beyond RAG for Agent Memory: Retrieval by Decoupling and Aggregation. [paper] [code] [project]
EMNLP 2025 Findings AI-Agent School (AAS): Evolution in Simulation: AI-Agent School with Dual Memory for High-Fidelity Educational Dynamics. [paper]
EMNLP 2025 ARIA: Enabling Self-Improving Agents to Learn at Test Time With Human-In-The-Loop Guidance. [paper] [code]
arXiv 2025 Decentralizing AI memory: SHIMI, a semantic hierarchical memory index for scalable agent reasoning. [paper]
REALM 2025 Workshop From knowledge to noise: CTIM-rover and the pitfalls of episodic memory in software engineering agents. [paper]
ICCV 2025 NavMorph: A Self-Evolving World Model for Vision-and-Language Navigation in Continuous Environments. [paper] [code]
EMNLP 2025 Findings PRINCIPLES: Synthetic Strategy Memory for Proactive Dialogue Agents. [paper] [code] [project]
NeurIPS 2025 Workshop SEDM: Scalable self-evolving distributed memory for agents. [paper]
arXiv 2025 SGMem: Sentence graph memory for long-term conversational agents. [paper]
ICLR 2024 Workshop OS-copilot: Towards generalist computer agents with self-improvement. [paper] [project]
NeurIPS 2024 Richelieu: Self-Evolving LLM-Based Agents for AI Diplomacy. [paper] [code]

Runtime Harness

ICLR 2026 Workshop AutoHarness: improving LLM agents by automatically synthesizing a code harness. [paper]
FSE 2026 AutonomyLens: AutonomyLens: A Self-Evolving Simulation-Based Testing Loop for Autonomous Systems. [paper]
arXiv 2026 Continual Harness: Continual Harness: Online Adaptation for Self-Improving Foundation Agents. [paper] [code]
ICML 2026 CUDAnalyst: Towards Feedback-to-Plan Decisions for Self-Evolving LLM Agents in CUDA Kernel Generation. [paper] [code]
arXiv 2026 Harness handbook: Making evolving agent harnesses readable,navigable, and editable. [paper] [project]
arXiv 2026 HarnessForge: Joint harness and policy evolution for adaptive agent systems. [paper] [code]
arXiv 2026 LedgerAgent: Structured state for policy-adherent tool-calling agents. [paper]
arXiv 2026 M*: M⋆: Every Task Deserves Its Own Memory Harness. [paper] [code] [project]
arXiv 2026 MemoHarness: Agent harnesses that learn from experience. [paper]
arXiv 2026 Meta-harness: End-to-end optimization of model harnesses. [paper]
arXiv 2026 Milkyway: Harnessing Pre-Resolution Signals for Future Prediction Agents. [paper]
arXiv 2026 Self-Harness: Harnesses That Improve Themselves. [paper]
arXiv 2026 SemaClaw: a step towards general-purpose personal AI agents through harness engineering. [paper]
arXiv 2026 The last harness you'll ever build. [paper]

Reliability and the Fixed-Improver Limit

arXiv 2026 391 Consecutive Sessions: Written by AI, Managed by AI: Semantic Space Control and Index Sickness Elimination Across 391 Consecutive Sessions. [paper] [code]
ACL 2026 Findings ErrorProbe: Towards Self-Improving Error Diagnosis in Multi-Agent Systems. [paper]
arXiv 2026 From Failed Trajectories to Reliable Agents: From Failed Trajectories to Reliable LLM Agents: Diagnosing and Repairing Harness Flaws. [paper] [code]
arXiv 2026 Harness Updating ≠ Benefit: Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents. [paper] [code]
ICML 2026 Workshop Library Drift: Library Drift: Diagnosing and Fixing a Silent Failure Mode in Self-Evolving LLM Skill Libraries. [paper] [code]
arXiv 2026 Ratchet: Ratchet: A Minimal Hygiene Recipe for Self-Evolving LLM Agents. [paper] [code]
arXiv 2026 RoboPhD: Evolving Diverse Complex Agents Under Tight Evaluation Budgets. [paper] [code]
arXiv 2026 SkillBrew: SkillBrew: Multi-Objective Curation of Skill Banks for LLM Agents. [paper]

🔁 L3: Improver-Level Self-Evolution _{↑ contents}

Deepest active evolution target: Improver. Characteristic failure: Metric capture.

Figure 6. The current improver helps produce or select a candidate successor to its own update mechanism. After external audit and promotion, the retained updater governs later proposal, selection, commit, and rollback under a fixed criterion.

↪️ Jump to: Self-Referential Agents (10) · Learning Better Improvement Strategies (7) · Reliability and the Fixed-Criterion Limit (4)

Self-Referential Agents

ICLR 2026 Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents. [paper] [code] [project]
arXiv 2026 Huxley-Gödel Machine: Human-Level Coding Agent Development by an Approximation of the Optimal Self-Improving Machine (HGM). [paper] [code]
arXiv 2026 HyperAgents: Self-Referential Agents that Integrate Task and Meta Agents. [paper] [code]
arXiv 2026 MOSS: Self-Evolution through Source-Level Rewriting in Autonomous Agent Systems. [paper] [code]
arXiv 2026 SePO: SePO: Self-Evolving Prompt Agent for System Prompt Optimization. [paper] [code]
ICLR 2025 Workshop A Self-Improving Coding Agent. [paper]
ACL 2025 Gödel Agent: A Self-Referential Agent Framework for Recursively Self-Improvement. [paper] [code]
arXiv 2025 Live-SWE-agent: Can Software Engineering Agents Self-Evolve on the Fly? [paper] [code] [project]
NAACL 2025 Self-Developing: Can Large Language Models Invent Algorithms to Improve Themselves?: Algorithm Discovery for Recursive Self-Improvement through Reinforcement Learning. [paper]
arXiv 2003 Gödel Machines: Self-Referential Universal Problem Solvers Making Provably Optimal Self-Improvements. [paper]

Learning Better Improvement Strategies

arXiv 2026 A-Evolve-Training: Autonomous Post-Training of a 30B Model. [paper] [code]
arXiv 2026 AREX: Towards a Recursively Self-Improving Agent for Deep Research. [paper]
arXiv 2026 EvoTrainer: Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learning. [paper] [code]
ACL 2026 Findings Polaris: A Gödel Agent Framework for Small Language Models through Experience-Abstracted Policy Repair. [paper]
arXiv 2026 Reinforcing Chain-of-Thought Reasoning with Self-Evolving Rubrics. [paper]
arXiv 2024 Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolution. [paper]
CoLM 2024 Self-Taught Optimizer (STOP): Recursively Self-Improving Code Generation. [paper] [code]

Reliability and the Fixed-Criterion Limit

arXiv 2026 ImprovEvolve: Basin-Hopping Meets LLM-Guided Evolutionary Search. [paper]
ICML 2026 Workshop Self-Supervised Theorem Discovery: Self-Supervised Theorem Discovery in a Formal Axiomatic System. [paper]
arXiv 2025 AlphaEvolve: A Coding Agent for Scientific and Algorithmic Discovery. [paper] [code] [project]
arXiv 2025 DeepEvolve: Scientific Algorithm Discovery by Augmenting AlphaEvolve with Deep Research. [paper] [code]

🎯 L4: Criterion-Level Self-Evolution _{↑ contents}

Deepest active evolution target: Criterion. Characteristic failure: Criterion drift.

↪️ Jump to: Evolving Evaluation Mechanisms (11) · Evolving Evaluation Tasks and Objectives (19)

Evolving Evaluation Mechanisms

ICML 2026 DR Tulu（RLER）: DR Tulu: Reinforcement Learning with Evolving Rubrics for Deep Research. [paper] [code] [project]
arXiv 2026 EvoRubric: EvoRubric: Self-Evolving Rubric-Driven RL for Open-Ended Generation. [paper]
EACL 2026 Findings Learning to Judge: LLMs Designing and Applying Evaluation Rubrics. [paper]
arXiv 2026 OpenComputer: Verifiable Software Worlds for Computer-Use Agents. [paper] [code]
arXiv 2026 Red Queen Gödel Machine: The Red Queen Gödel Machine: Co-Evolving Agents and Their Evaluators. [paper]
ACL 2026 RubricBench: Aligning Model-Generated Rubrics with Human Standards. [paper]
arXiv 2026 SkillCoach: SkillCoach: Self-Evolving Rubrics for Evaluating and Enhancing Agentic Skill-Use. [paper]
arXiv 2026 Who grades the grader? Co-evolving evaluation metrics and skills for self-improving LLM agents. [paper] [code]
EMNLP 2025 Meta-Rewarding Language Models: Self-Improving Alignment with LLM-as-a-Meta-Judge. [paper]
COLING 2024 Calibrating LLM-Based Evaluator. [paper]
arXiv 2024 Self-taught evaluators. [paper]

Evolving Evaluation Tasks and Objectives

ICLR 2026 AC/DC: Discovering Novel LLM Experts via Task-Capability Coevolution. [paper] [code] [project]
ICLR 2026 (Oral) AdAEM: An Adaptively and Automated Extensible Measurement of LLMs' Value Difference. [paper]
TACL 2026 Learning to Optimize Multi-Objective Alignment Through Dynamic Reward Weighting. [paper] [project]
Knowledge-Based Systems 2025 A Large Language Model-Driven Reward Design Framework via Dynamic Feedback for Reinforcement Learning. [paper]
ICLR 2025 AutoBencher: Towards declarative benchmark construction. [paper]
AAMAS 2025 DyLam: A Dynamic Reward Weighting Framework for Reinforcement Learning Algorithms. [paper]
NAACL 2025 High-Dimension Human Value Representation in Large Language Models. [paper]
ICML 2025 R*: Efficient Reward Design via Reward Structure Evolution and Parameter Alignment Optimization with Large Language Models. [paper]
NeurIPS 2024 ALI-Agent: Assessing LLMs' Alignment with Human Values via Agent-Based Evaluation. [paper]
NeurIPS 2024 CLAVE: An Adaptive Framework for Evaluating Values of LLM Generated Responses. [paper]
ICLR 2024 Eureka: Human-Level Reward Design via Coding Large Language Models. [paper] [code] [project]
NeurIPS 2024 EvoCodeBench: An Evolving Code Generation Benchmark with Domain-Specific Evaluations. [paper]
NeurIPS 2024 (Spotlight) ProgressGym: Alignment with a Millennium of Moral Progress. [paper] [code]
ICLR 2024 Text2Reward: Reward shaping with language models for reinforcement learning. [paper] [project]
AAAI 2024 Value Kaleidoscope: Engaging AI with Pluralistic Human Values, Rights, and Duties. [paper]
ICML 2022 Evolving curricula with regret-based environment design. [paper]
NAACL 2021 Dynabench: Rethinking benchmarking in NLP. [paper]
NeurIPS 2020 Emergent Complexity and Zero-shot Transfer via Unsupervised Environment Design. [paper] [code] [project]
arXiv 2019 Paired Open-Ended Trailblazer (POET): Endlessly Generating Increasingly Complex and Diverse Learning Environments and Their Solutions. [paper] [code] [project]

🛡️ Cross-Level Reliability: Evidence, Acceptance, and Control _{↑ contents}

Works supporting the survey's cross-level synthesis of shared failure modes, external audit signals, promotion reliability, and audit independence under compromise.

The reliability ladder for self-evolving agents

Figure 7. Each step is a deeper evolution target, and each card pairs it with the external audit and the control that a claim at that depth requires. The rise of the steps is self-evolution depth, not capability or reliability.

↪️ Jump to: External Audit Across Self-Evolution Levels (3) · Level-Specific Audit Failures and Evaluation Horizons (3) · Preserving Audit Independence Under Compromise (2)

External Audit Across Self-Evolution Levels

arXiv 2026 Anytime-Valid Certificates: Self-Evolving Agents with Anytime-Valid Certificates. [paper]
arXiv 2026 Held-Out Selection: Recursive Self-Evolving Agents via Held-Out Selection. [paper]
arXiv 2026 PACE (Acceptance Tests): PACE: Anytime-Valid Acceptance Tests for Self-Evolving Agents. [paper]

Level-Specific Audit Failures and Evaluation Horizons

arXiv 2026 Generative-Evaluative Agreement: Generative-Evaluative Agreement: A Necessary Validity Criterion for LLM-Enabled Adaptive Assessment. [paper]
ICML 2026 Not Always Faithful Self-Evolvers: Large Language Model Agents Are Not Always Faithful Self-Evolvers. [paper] [code]
ICML 2026 Workshop SEVA: SEVA: Self-Evolving Verification Agent with Process Reward for Fact Attribution. [paper] [code]

Preserving Audit Independence Under Compromise

arXiv 2026 Safety in Self-Evolving Systems: Safety in Self-Evolving LLM Agent Systems: Threats, Amplification, and Case Studies. [paper]
ICLR 2026 Workshop Zombie Agents: Zombie Agents: Persistent Control of Self-Evolving LLM Agents via Self-Reinforcing Injections. [paper]

🚀 Open Problems and Outlook _{↑ contents}

Works cited in the research agenda on capability growth, longitudinal and adaptive evaluation, deployment, and scalable oversight.

Applications and staged deployment of self-evolving agents

Figure 8. Across executable engineering, persistent digital agents, scientific discovery, and embodied or high-stakes systems, the available evidence ranges from executable checks to expert review and backtests. Wider deployment requires staged evaluation against a declared external target.

↪️ Jump to: Evolution: Capability Growth and Learning Over Time (2) · Evaluation: Longitudinal Measurement and Adaptive Auditing (22) · Applications: From Updates to Deployment (5) · Governance: Goal Preservation and Scalable Oversight (1)

Evolution: Capability Growth and Learning Over Time

arXiv 2026 AlphaZero for Connect Four: Frontier Coding Agents Can Now Implement an AlphaZero Self-Play Machine Learning Pipeline For Connect Four That Performs Comparably to an External Solver. [paper]
arXiv 2026 Computational Boundary of Inference: The Computational Boundary of Inference: Capability Internalization, Training, and the Turing Jump. [paper]

Evaluation: Longitudinal Measurement and Adaptive Auditing

arXiv 2026 Agent2 RL-Bench: Can LLM Agents Engineer Agentic RL Post-Training? [paper]
ACL 2026 Findings AgenticEval: AgenticEval: Toward Agentic and Self-Evolving Safety Evaluation of Large Language Models. [paper]
arXiv 2026 BenchTrace: BenchTrace: A Benchmark for Testing Reflection Ability and Controlled Evolution in LLM Agents. [paper] [code]
arXiv 2026 Capability→Subjective Behavior: Does Capability Transfer to Subjective Behavior — and Would Our Instruments Tell Us? A Self-Evolving, Trust-by-Construction Evaluation Paradigm. [paper]
arXiv 2026 ClawArena: ClawArena: Benchmarking AI Agents in Evolving Information Environments. [paper] [code]
arXiv 2026 Elmes*: ELMES+: Automated Construction of Fine-Grained Evaluation Rubrics for Large Language Models in Long-Tail Educational Scenarios. [paper]
arXiv 2026 EvoMemBench: EvoMemBench: Benchmarking Agent Memory from a Self-Evolving Perspective. [paper] [code]
arXiv 2026 Frontier-Eng: Frontier-Eng: Benchmarking Self-Evolving Agents on Real-World Engineering Tasks with Generative Optimization. [paper] [code] [project]
arXiv 2026 MathDuels: MathDuels: Evaluating LLMs as Problem Posers and Solvers. [paper] [project]
arXiv 2026 Meta-Agent Challenge: The Meta-Agent Challenge: Are Current Agents Capable of Autonomous Agent Development? [paper] [code] [project]
arXiv 2026 PAST-Bench: Benchmarking the Foundations of Recursive Self-Improvement in Personal Agents. [paper]
arXiv 2026 PixJail: PIXJAIL: Self-Evolving Paper-to-Pipeline Reproduction for Text-to-Image Jailbreak Evaluation. [paper]
arXiv 2026 PostTrainBench: Can LLM Agents Automate LLM Post-Training? [paper]
arXiv 2026 RSIBench-Data: Benchmarking Data-Centric Research for Recursive Self-Improvement. [paper]
arXiv 2026 SAGE (Socialized Evolution): SAGE: A Quantitative Evaluation of Socialized Evolution in Agent Ecosystems. [paper]
arXiv 2026 SE-Bench: SE-BENCH: Benchmarking Self-Evolution with Knowledge Internalization. [paper] [code]
arXiv 2026 SEA-Eval: SEA-Eval: A Benchmark for Evaluating Self-Evolving Agents Beyond Episodic Assessment. [paper]
arXiv 2026 SEAGym: SEAGym: An Evaluation Environment for Self-Evolving LLM Agents. [paper]
arXiv 2026 SrDetection: SrDetection: A Self-Referential Framework for Data Leakage Detection in Code Large Language Models. [paper]
ICML 2025 GETA: Raising the Bar: Investigating the Values of Large Language Models via Generative Evolving Testing. [paper]
ACL 2024 I am a Strange Dataset: I am a Strange Dataset: Metalinguistic Tests for Language Models. [paper] [code]
ICLR 2024 LAMEN: Evaluating Language Model Agency Through Negotiations. [paper] [code]

Applications: From Updates to Deployment

arXiv 2026 ENPIRE: Agentic robot policy self-improvement in the real world. [paper]
ICLR 2026 FactorMiner: a self-evolving agent with skills and experience memory for financial alpha discovery. [paper]
arXiv 2025 Enabling autonomic microservice management through self-learning agents. [paper]
EMNLP 2025 Investigating pedagogical teacher and student LLM agents: Genetic adaptation meets retrieval-augmented generation across learning styles. [paper]
NeurIPS 2025 Workshop Mobile-agent-e: Self-evolving mobile assistant for complex tasks. [paper] [project]

Governance: Goal Preservation and Scalable Oversight

Microsoft Research 2026 Agentic evolution: From self-improving agents to co-evolving human–AI systems. [paper]

⚖️ License _{↑ contents}

Original text and images in this repository are licensed under the MIT License. Linked papers, code repositories, project pages, names, and third-party metadata remain subject to their respective copyright and license terms.