Co-Evolving Harnesses and Models: On-Policy Correction Helps Weaker Models Catch Up Where Imitation Fails
- URL: http://arxiv.org/abs/2609.09134v1
- Date: Tue, 08 Sep 2026 17:53:49 GMT
- Title: Co-Evolving Harnesses and Models: On-Policy Correction Helps Weaker Models Catch Up Where Imitation Fails
- Abstract summary: Automated harness evolution can enable smaller models to perform well on domain-specific tasks at a fraction of frontier-model cost.<n>We show that imitation transfers knowledge and increases scaffold usage, but disrupts model-harness fit.<n>We develop an on-policy expert-correction pipeline, automated by a meta-level MLE agent, that localizes the failing turn in the weaker model's own rollout and asks the expert to rewrite only that turn.
- Score: 23.287109132562403
- License: http://creativecommons.org/publicdomain/zero/1.0/
- Abstract: Agent harnesses (the system prompt, tool set, execution hooks, and context-management scaffolding around a model) are a critical determinant of agentic task success. Automated harness evolution can enable smaller models to perform well on domain-specific tasks at a fraction of frontier-model cost. Since both the harness and model weights shape behavior, we ask how harness evolution and lightweight fine-tuning should be combined. Across seven enterprise agent tasks, we first evolve a harness with the weaker model, then find that a stronger expert often uses it more effectively, suggesting expert supervision could close the remaining gap. However, training the weaker model on the expert's complete trajectories under the evolved harness backfires: performance regresses on all seven tasks by 4 to 30 points across Qwen3-Coder and Gemma 4, even though the same procedure helps under the unevolved harness. Our analysis shows that imitation transfers knowledge and increases scaffold usage, but disrupts model-harness fit: the weaker model adopts the expert's planning strategy without the competence to execute it and no longer matches the harness evolved around its native planning style. We therefore develop an on-policy expert-correction pipeline, automated by a meta-level MLE agent, that localizes the failing turn in the weaker model's own rollout and asks the expert to rewrite only that turn. This preserves the model's planning style and combines the gains of harness evolution and model adaptation. Our results identify and resolve a source of contention between harness and weight updates, yielding a compatibility-preserving recipe for economical co-evolution on domain-specific enterprise tasks.
Related papers
- EVOHARNESSBENCH: Can Your Agents Keep Pace with an Evolving Harness? [83.21711423118768]
EVOHARNESSBENCH is a benchmark for evaluating agents under controlled harness evolution.<n>It contains 17 multi-stage harness streams constructed deterministically from verifier-based benchmarks.
arXiv Detail & Related papers (2026-09-03T03:24:02Z) - HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness? [61.02699867449589]
We introduce HarnessDev, a benchmark that shifts the unit of evaluation from task outputs to run infrastructure.<n>In Creation, the agent starts from a minimal seed and a small number of cases, then builds a complete execution system.<n>In Evolution, it starts from its own created harness and iteratively revises it using downstream execution feedback, with the goal of improving benchmark performance.
arXiv Detail & Related papers (2026-09-01T15:45:33Z) - JIT-Agent: Scaling Harness Intelligence via Just-in-Time Harness Evolution [74.58840188662151]
JIT-Agent is a harness intelligence model trained to synthesize task-adaptive agent harnesses on the fly.<n>JIT-Agent is the first model purpose-built for just-in-time harness generation.
arXiv Detail & Related papers (2026-08-26T10:05:33Z) - HELIX: Model-Harness Co-evolution for Recursive Self-Improvement [8.484282837156933]
We present HELIX, a source-traceable substrate for harness evolution.<n>We show how harness evolution expands current capability and creates learning signal for the next model.
arXiv Detail & Related papers (2026-08-14T04:44:53Z) - AI4AI at Test-Time: Strong-to-Weak Capability Transfer via Harnesses [72.0777620499307]
We study whether a stronger builder model can construct inference-time harnesses that help a weaker target model solve tasks more reliably without any parameter updates.<n> Empirically, this form of test-time capability transfer is highly effective, nearly doubling average target-model performance from 0.49 to 0.91.<n>We find that builder-model reasoning effort improves harness quality monotonically, platform effects are modest relative to the builder model's own capability, and weaker target models receive the largest gains.
arXiv Detail & Related papers (2026-08-12T17:53:18Z) - Self-Evolving Embodied Agents via Skill-Harness Evolution [53.307052568223945]
Embodied agents are increasingly built as systems around foundation models.<n>While supervised fine-tuning and reinforcement learning can adapt agents to new environments, they require additional data, rewards, and training runs.<n>We propose SHAPER, a self-evolving framework for train-free embodied adaptation that keeps model parameters frozen.
arXiv Detail & Related papers (2026-08-11T18:55:58Z) - HarnessX: A Composable, Adaptive, and Evolvable Agent Harness Foundry [35.87794858139959]
HarnessX is a foundry for composable, adaptive, and evolvable agent harnesses.<n>It assembles typed harness primitives via a substitution algebra, adapts them through AEGIS.<n>It closes the harness-model loop by turning trajectories into both harness updates and model training signal.
arXiv Detail & Related papers (2026-06-12T08:27:11Z) - Self-Harness: Harnesses That Improve Themselves [30.839060071524433]
Self-Harness is a new paradigm in which an LLM-based agent improves its own operating harness without relying on human engineers or stronger external agents.<n>We instantiate Self-Harness on Terminal-Bench-2.0 using a minimal initial harness and three base models from diverse families.
arXiv Detail & Related papers (2026-06-08T13:50:23Z) - Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents [82.27610290890475]
We analyze two harness self-evolution capabilities: (i) harness-updating, the capability to produce useful persistent harness updates from execution evidence; (ii) harness-benefit, the capability to benefit from updated harnesses during task solving.<n>Our analysis reveals two findings. First, harness-updating is flat in base capability: models from different capability tiers produce harness updates that lead to surprisingly similar gains.<n>Second, harness-benefit is non-monotonic in base capability: weak-tier models benefit little from updated harnesses, mid-tier models benefit most, and strong-tier models benefit less than mid-tier.
arXiv Detail & Related papers (2026-05-28T22:16:14Z) - DemoEvolve: Overcoming Sparse Feedback in Agentic Harness Evolution with Demonstrations [12.048713464721665]
DemoEvolve is a demonstration-bootstrapped approach to harness evolution.<n>We study this paradigm as a form of sample-efficient fast adaptation.<n>We show that DemoEvolve produces more effective and auditable harness edits.
arXiv Detail & Related papers (2026-05-23T12:10:52Z)
This list is automatically generated from the titles and abstracts of the papers in this site.
This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.