Growing Harness Turns Task Feedback into Reusable Agent Code
Grow the Harness, Not the Context: From Strategy-Free Scaffolds to Reusable Specialist Agents
Moving recurring control out of context and into code means smaller models can run competent agents at a fraction of the inference cost, and the grown harness accumulates capability as reusable specialist code rather than prompts. Teams building agents on cost-sensitive or small-model deployments can apply this failure-guided repair loop directly, especially where Tool-Calling baselines collapse at the 4B scale. The ablations on trace-local edits, joint repair and gate rollback give a concrete recipe for anyone attempting similar harness-growing systems.
A paper on arXiv introduces Growing Harness, a failure-guided training paradigm that converts recurring agent control decisions into reusable executable code instead of keeping them in model context.
It starts from a strategy-free scaffold with fixed model and tool interfaces, localises failures via function-level execution traces, and uses joint repair plus a success-first held-out gate to roll back harmful edits.
Across BrowseComp-Plus and WebArena-Verified with deployment models from 4B to 120B parameters, it achieves the highest mean success in five of six benchmark-model settings and trails the best mean by 0.
7 pp in the sixth.Compared with a Tool-Calling agent, it cuts LLM calls by 76.
0-91.8% and inference cost by 74.
4-98.6%.
On WebArena-Verified its success stays at 44.7-45.
3% across model scales, while Tool-Calling drops to 6.7% with the 4B model.
Grow the Harness, Not the Context: From Strategy-Free Scaffolds to Reusable Specialist Agents
Large language model (LLM) agents often handle streams of related tasks, yet standard harnesses repeatedly ask the model to reconstruct the same control decisions inside each task's context. We study whether task feedback can instead turn recurring control into reusable executable code, while reserving LLM calls for task-specific semantic reasoning. We introduce Growing Harness, a failure-guided training paradigm that learns the agent harness itself from a strategy-free scaffold that exposes fixed model and tool interfaces but encodes no task-solving controller. Function-level execution traces localize each failure to a bounded code surface, an optimizer repairs a window of failures jointly, and a success-first held-out gate rolls back repair sequences that harm prior capability. Accepted edits accumulate in one shared harness, allowing its control structure to emerge from task feedback. Across BrowseComp-Plus and WebArena-Verified with three deployment models from 4B to 120B parameters, Growing Harness achieves the highest mean success in five of six benchmark-model settings and trails the best mean by 0.7 pp. in the sixth. Relative to a Tool-Calling agent, it reduces LLM calls by 76.0-91.8% and deployed-agent inference cost by 74.4-98.6%. On WebArena-Verified, its success remains 44.7-45.3% across model scales, whereas Tool-Calling falls to 6.7% with the 4B model. Ablations show that trace-local edits, joint repair, and gate-based rollback each improve final success. These results show that persistent program growth can move recurring control out of model context and into low-cost code, yielding reusable specialist agents that remain effective with smaller deployment models.
This report is categorized under AITOP's "Papers" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Growing Harness, LLM, agents, inference cost.
Source attribution: curated from arXiv cs.AI (@Laizhen Li, Jiarui Li, Juanjuan Zhao, Kejiang Ye, Ye Li, Cheng-zhong Xu, Xitong Gao). Cross-source corroboration: 2 independent sources have covered this development (including elvis), confirming industry-wide signal strength.
- elvis09-21 23:00Source
