Reward engine guides tool agents to improve task success
Reward Engine: Reward-Centric Self-Evolution for Open-Ended Tool Agents
Artificial Intelligence
Summary
It is hard for AI agents that use different tools to know how well they are doing because there is no single way to tell if they succeeded. The authors created a method called Reward Engine that first figures out what success looks like by building tests and quality checks from examples. Then, it uses these tests like a teacher to help the agent improve its actions and learn better. This method helps AI agents get more consistent and higher quality results when completing tasks.
What this means in practice
- •For software developers: Help automate testing and improvement of AI tasks that have multiple solution paths without explicit rewards.
- •For automated customer support teams: Use Reward Engine to refine AI tool workflows by automatically judging and improving task success beyond fixed scripts.
Authors
Litong Feng, Ji Qi, Yiran Cheng, Shaowu Pan, Xinyu Zhang, Qijun Li
Abstract
Open-ended tool agents can complete the same objective through different tool paths, and their quality depends jointly on the interaction trace, environment state, and final artifacts, so such tasks rarely admit an executable ground-truth reward. The central challenge is therefore to construct a reward from existing experience that exposes the current policy's failures while remaining stationary during optimization. We introduce Reward Engine, which, like a self-motivated student, first works out what success requires and then optimizes toward it. From the task and initial trajectories, it builds a RewardPack of testcases for required correctness and rubrics for graded quality, each executed as a check that returns a judgment with evidence. The frozen RewardPack then guides the agent to revise and select its outputs, and can also serve as a reward for reinforcement learning. On WorkBuddy Bench Office and Claw-Eval, the constructed reward largely agrees with the benchmark graders, not reversing their ordering on 82.3\% and 80.5\% of candidate pairs. Used to revise executions, it raises three-run consistency ($\mathrm{Pass}^{3}$) over Baseline from 45.0\% to 54.0\% and from 78.4\% to 83.7\%.