WEFT (Whole-system Evolution For Tool-use Post-training) scales agentic capabilities by coupling environment breadth and task complexity with execution-driven self-evolution. It moves beyond scaling isolated environments to ensure coherent interaction between the agent, task, and evaluator.
HOW THIS AFFECTS YOU
●
builderThis method offers a more robust way to train agents that can reliably interact with complex tool ecosystems.
●
researcherYou can explore how whole-system evolution improves learning signals compared to isolated environment scaling.