Harness-of-Harness: Multi-Day Autonomous Software Development with Continual Improvement
A paper introduces Harness-of-Harness (HoH), a framework that enables LLM-based coding agents to continually improve software during autonomous development. HoH organizes existing coding-agent harness executions into iterative planning-coding-testing loops, balances repair with capability growth, scopes development into small verifiable increments, separates implementation-time testing from independent evaluation, and constrains verifiable outputs. On GameCraft-Bench, FrontierSWE, and ProgramBench, three harness-model pairs (Codex with GPT-5.5, OpenCode with DeepSeek-V4-Pro, and Pi with MiniMax-M3), HoH consistently outperforms standalone harnesses, achieving an average relative gain of 52.25 percent.
The paper studies autonomous software development where LLM-based coding agents transform high-level requirements into complete, functional software systems without human intervention. Harness-of-Harness (HoH) is introduced as a framework that enables coding agents to continually improve software during autonomous development. HoH operates on existing coding-agent harnesses and organizes their executions into iterative planning-coding-testing loops. To sustain improvement across loops, HoH balances repair with capability growth, scopes development into small and verifiable increments, separates implementation-time testing from independent evaluation, and constrains verifiable outputs rather than prescribing agent workflows. It progressively exposes deliverables, role-specific tools, and skills, encourages reuse rather than recreation, and maintains versioned project histories. On GameCraft-Bench, FrontierSWE, and ProgramBench, three harness-model pairs (Codex with GPT-5.5, OpenCode with DeepSeek-V4-Pro, and Pi with MiniMax-M3), HoH consistently outperforms the corresponding standalone harnesses, achieving an average relative gain of 52.25 percent and a maximum gain (truncated in evidence).
HoH introduces a meta-harness that wraps existing coding-agent harnesses into iterative loops, emphasizing verifiable increments and independent evaluation to avoid compounding errors. The reported average relative gain of 52.25% across three benchmarks suggests that continual improvement mechanisms can substantially boost autonomous coding performance. Observable next signals include publication of full benchmark results, open-source release of HoH, and independent replication on additional harness-model pairs.
This research signals a shift toward multi-day autonomous software development, where agents not only generate code but also maintain and improve it over time. The framework's focus on reuse, versioned histories, and role-specific tools aligns with enterprise needs for maintainable and auditable AI-generated software. Adoption may accelerate as coding agents move from single-shot tasks to long-running projects.
HoH's approach could reduce development costs and time-to-market by enabling coding agents to self-improve and handle larger scopes autonomously. Enterprises may benefit from more reliable AI-generated software with versioned histories and independent evaluation, lowering risk and increasing trust in autonomous development pipelines.
If HoH's gains generalize, autonomous coding agents could handle increasingly complex, multi-day development tasks with less human oversight. Near-term signals to watch include integration of HoH-like loops into commercial coding tools, benchmark expansions, and evidence of reduced human intervention in real-world projects. Longer term, this could contribute to fully autonomous software teams.