Self-Improving Agents Need Harness Edits Tested on Fresh Tasks
A strong small model with an evolved harness can outscore a larger model tuned only with skills, shifting attention from model choice to harness quality. Builders running self-modifying agents should gate every edit behind fresh-task evaluation and keep evaluators, permissions and emergency stops out of the agent's reach. The taxonomy of edits, from prompt changes to runtime code, gives teams a concrete checklist for admission policies.
A method maps harness self-improvement for agents onto evolutionary algorithms, ranging from keep-the-best search to archives, Pareto selection and bandit-guided budgets. The harness covers the prompts, memory, tools and code wrapped around a fixed model, and accepted edits carry into later runs. The admission rule keeps an edit only if it beats the current harness on new, independent tasks by a set margin. In a compiled case from prior work, GPT-5.4-nano with an evolved harness and skills scored 77.58% on SpreadsheetBench, versus 76.20% for GPT-5.5 with skill tuning alone, though costs were not matched.
Self-improving agents should accept a harness edit only after it beats the current version on fresh, independent tasks.
It maps harness self-improvement methods onto evolutionary algorithms, from simple keep-the-best search to archives, Pareto selection, and bandit-guided budgets.
Self-improving agents now rewrite their harness: the prompts, memory, tools, and code around a fixed model. Accepted edits carry into later runs, so a bad edit can overfit, erase old skills, or spread to other users.
They sort these edits by what they change, from prompts to runtime code. Their admission rule: keep an edit only if it beats the current harness on new, independent tasks by a set margin.
In a case they compile from prior work, GPT-5.4-nano with an evolved harness and skills scored 77.58% on SpreadsheetBench. GPT-5.5 with skill tuning alone scored 76.20%, though costs were not matched.
If your agent edits its own setup, test each change on fresh tasks, and keep evaluators, permissions, and emergency stops out of its reach.
