mlsanityData to 14 Oct 2026

Where Does Harness-Optimization Value Live? Localized Gains and the Budget-Splitting Trap in Self-Evolving LLM Agents

arXiv 2609.02889 ↗First posted 25 Jun 2026cs.CLcited by 1

On the map: LLM agents and tool use → Software engineering coding agents

also in 1 more
Abstract (arXiv)

A growing body of work improves frozen large language models (LLMs) as agents by evolving their harness: the textual scaffolding around the model, including persona, strategy, format rules, and control heuristics. Existing reflective prompt-evolution methods usually optimize this harness as one flat string. We instead ask where the optimization value actually resides. We introduce HARNESSEVO, which decomposes the harness into four separately evolvable slots: role, task-strategy, tool/format-rules, and reflection/control. Using the same reflective optimizer under an iso-budget setting, we pair this decomposition with leave-one-in and leave-one-out attribution to measure the contribution of each slot. On ALFWorld with a frozen 7B backbone, HARNESSEVO does not significantly improve the overall binary success rate over either the stock harness or flat-string evolution: 0.657 versus 0.642 and 0.642, respectively. However, the slot-level analysis reveals that nearly all useful optimization value is localized in the reflection/control slot, which achieves a leave-one-in gain of +0.119. The other slots are individually null. We further show that uniform budget splitting is harmful: allocating 64 rollouts across four slots leaves only 16 per slot, below the optimizer's effective search floor, causing every slot to freeze at its empty seed. Concentrating the budget on the high-credit control slot recovers the lost gain, reaching 0.761 with half the split budget. The effect is task-contingent. On WebShop, all slots freeze empty and all methods tie, indicating a genuine absence of recurrent, verbalizable control failures rather than budget starvation. Overall, our results suggest that harness value is localized, uniform budget splitting can be actively harmful, and credit assignment should precede structured agent-evolution.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Software engineering coding agents, one of 16 directions in LLM agents and tool use. Builds on 63 of its 66 references that are in the map. First posted on arXiv 25 Jun 2026.

Builds on

63 of 66 references.

References · 63

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,297
  3. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,695
  4. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  5. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  6. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
  7. 2023Reflexion: Language Agents with Verbal Reinforcement Learningcited by 4,276
  8. 2023Generative Agents: Interactive Simulacra of Human Behaviorcited by 3,568
  9. 2023Self-Refine: Iterative Refinement with Self-Feedbackcited by 3,442
  10. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,373
  11. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Modelscited by 3,211
  12. 2020Learning to summarize from human feedbackcited by 2,588
38 more
  1. 2023G-Eval: NLG Evaluation using GPT-4 with Better Human Alignmentcited by 2,276
  2. 2023Voyager: An Open-Ended Embodied Agent with Large Language Modelscited by 2,191
  3. 2020REALM: Retrieval-Augmented Language Model Pre-Trainingcited by 2,141
  4. 2023WebArena: A Realistic Web Environment for Building Autonomous Agentscited by 2,006
  5. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,986
  6. 2023ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIscited by 1,947
  7. 2023Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflectioncited by 1,839
  8. 2016Concrete Problems in AI Safetycited by 1,642
  9. 2024A Survey on LLM-as-a-Judgecited by 1,423
  10. 2024τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domainscited by 1,402
  11. 2023MemGPT: Towards LLMs as Operating Systemscited by 1,389
  12. 2025s1: Simple test-time scalingcited by 1,341
  13. 2023AgentBench: Evaluating LLMs as Agentscited by 1,267
  14. 2022WebShop: Towards Scalable Real-World Web Interaction with Grounded Language Agentscited by 1,251
  15. 2020ALFWorld: Aligning Text and Embodied Environments for Interactive Learningcited by 1,205
  16. 2023GAIA: a benchmark for General AI Assistantscited by 1,128
  17. 2022Scaling Laws for Reward Model Overoptimizationcited by 1,095
  18. 2022Large Language Models Are Human-Level Prompt Engineerscited by 1,045
  19. 2023Large Language Models Cannot Self-Correct Reasoning Yetcited by 1,017
  20. 2022STaR: Bootstrapping Reasoning With Reasoningcited by 940
  21. 2023ExpeL: LLM Agents Are Experiential Learnerscited by 876
  22. 2023Large Language Models as Optimizerscited by 864
  23. 2023DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelinescited by 832
  24. 2023CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquingcited by 783
  25. 2022Transformers learn in-context by gradient descentcited by 732
  26. 2024LLM Evaluators Recognize and Favor Their Own Generationscited by 686
  27. 2022What learning algorithm is in-context learning? Investigations with linear modelscited by 684
  28. 2024Self-Rewarding Language Modelscited by 681
  29. 2022Large Language Models Can Self-Improvecited by 664
  30. 2023RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedbackcited by 625
  31. 2023Prometheus: Inducing Fine-grained Evaluation Capability in Language Modelscited by 575
  32. 2023MemoryBank: Enhancing Large Language Models with Long-Term Memorycited by 556
  33. 2023Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolutioncited by 551
  34. 2025GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learningcited by 542
  35. 2024From Generation to Judgment: Opportunities and Challenges of LLM-as-a-judgecited by 494
  36. 2025Agentic Context Engineering: Evolving Contexts for Self-Improving Language Modelscited by 398
  37. 2024Agent Workflow Memorycited by 383
  38. 2022The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Modelscited by 328

50 most cited of 63 All 63 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Harness Compilation: Which Decisions Should a Small Vision-Language Model Keep?cited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026HarnessCompass: Guiding Automatic Harness Evolution toward Generalizable and Effective Agent Harnesses
  2. 2026DemoEvolve: Overcoming Sparse Feedback in Agentic Harness Evolution with Demonstrations
  3. 2026Harness Evolution as Learning: Approximation, Generalization, and Optimization Limits of Self-Improving Personal Agents
  4. 2026Evo-Bench: Can Language Models Improve Agent Harness?
  5. 2026Evo2: Evolving the Evolver through Weight-Harness Co-Evolution
  6. 2026PACE: Two-Timescale Self-Evolution for Small Language Model Agents
  7. 2026What Makes an LLM a Good Optimizer? A Trajectory Analysis of LLM-Guided Evolutionary Search
  8. 2026Rethinking the Evaluation of Harness Evolution for Agents