Weft improves tool-use learning for AI by refining whole systems
WEFT: Scaling Tool-Use Post-Training for General-Purpose Agents
Artificial Intelligence
Summary
Teaching AI agents to use tools well is tricky because they need to interact smoothly with environments, tasks, and evaluators all at once. The authors found that focusing only on improving environments isn’t enough to boost performance. They created WEFT, a new method that helps the entire system work better by spotting and fixing failures continuously and keeping training stable. Tests showed that AI models using WEFT learned to complete tool-based tasks more effectively across many challenges.
What this means in practice
- •For ai system developers: Build AI agents that learn complex tool-using behaviors more reliably by improving whole-system feedback and training stability.
- •For workflow automation teams: Enhance AI tools that execute long and complex sequences of tasks by reducing failures through iterative system-level revision.
Authors
Bo Mao, Hang He, Linting Wang, Lizhi Lin, Maosen Zhou, Guanming Liu, Jinxiu Liu, Tianyu Huai, Chaoyun Zhang, Bingxuan Li, Kepeng Lei, Guanting Dong, Zhou Shao, Rui Zheng, Hang Yan, Jie Zhou, Chengcheng Wan, Tao Gui, Liang He, Xipeng Qiu
Abstract
Recent efforts to scale tool-use post-training have largely centered on the synthesis of executable environments, which constitute only one component of a broader agentic interaction system comprising the environment, task, agent harness, and evaluator. Scaling environments in isolation, however, does not guarantee commensurate gains in model performance, because reliable learning signals depend on coherent interactions among all components of the agentic interaction system. To address this problem, we introduce WEFT (Whole-system Evolution For Tool-use Post-training), which couples scalable agentic interaction system construction, execution-driven self-evolution, and stable post-training. WEFT scales agentic interaction system construction across environment breadth, task complexity, and interaction diversity. Execution-driven self-evolution iteratively uses execution traces and state evidence to attribute failures and revise the responsible components, with fresh rollouts evaluating the changes and providing evidence for subsequent evolution rounds. For stable post-training at scale, WEFT addresses both optimization and execution reliability: prefix-preserving sampling retains verified progress and atomic-turn credit assignment localizes learning signals, while MegaMCP maintains isolated, recoverable state across concurrent rollouts over shared tool services. Extensive experiments across various models and benchmarks demonstrate the effectiveness of WEFT for tool-use post-training. WEFT-8B and WEFT-14B outperform all evaluated matched-size environment-scaling baselines on BFCL V4, $τ^2$-Bench, and Claw-Eval. In particular, WEFT-14B improves over Agent-World-14B by 6.41, 2.23, and 12.27 percentage points. WEFT-35B-A3B further extends these gains to more challenging long-horizon workflow benchmarks, including Toolathlon-Verified and AutomationBench.