LiteEvo improves AI agent task performance with cheaper evolution
LiteEvo: Automated, Cost-Efficient Harness Evolution for Generalization to Unseen Tasks
Artificial Intelligence
Summary
AI systems often need special setups, called harnesses, to work well on tasks. The authors created LiteEvo, a method that automatically builds these setups without human help, starting from a simple common base. This method reuses learned parts from past runs and adapts to new tasks more efficiently and cheaply than earlier methods. LiteEvo was tested on multiple benchmark tasks and showed better performance and cost savings while also working well on tasks it hadn’t seen before.
What this means in practice
- •For ai system developers: Develop automated harnesses that improve AI agent task success while reducing computational and API token costs.
- •For software engineering teams: Create reusable AI agent components that maintain effectiveness across unseen tasks, aiding scalable and adaptable AI solutions.
Authors
Euntae Choi, Sumin Song, Sungjoo Yoo
Abstract
An LLM agent is defined by two things: the weights inside its model and the harness of components assembled around it. Harnesses are still handcrafted, and HarnessX, which evolves them automatically, starts each benchmark from a handcrafted harness, reports gains on the tasks it evolved on, and budgets 100 to 175 million meta-agent tokens per benchmark. We propose LiteEvo, a lightweight harness-evolution algorithm whose tool-free meta-agents mine agent trajectories for reusable components, curate them into a versioned library, and compose each round's harness from it, starting every benchmark from the same neutral harness and never naming the benchmark. Evolving on the graded tasks of five agentic benchmarks with a frozen Qwen3.5-9B, LiteEvo lifts pass@2 by 10.5 to 67.7pp and reaches comparable or higher pass@2 than a reproduction of HarnessX (71.0 against 67.3 on average) at 13.0 lower mean API cost. Harnesses evolved on train tasks keep their gains on unseen test tasks of four benchmarks, and LiteEvo also lifts Claude Code with Sonnet 4.6 by 1.2 to 71.4pp.