Build Hour: Agent RFT

In this Build Hour, Christine introduces Will and Theo, who delve into Agent RFT (Reinforcement Fine-Tuning), explaining how it enhances agent performance by allowing models to interact with the outside world through tools during training. They discuss optimizing agents through prompt engineering, task simplification, and tool improvement before considering fine-tuning. Will details Agent RFT's benefits, including improved reasoning, tool usage, and sample efficiency, while Theo emphasizes reduced latency and the importance of mirroring production environments for tool calls and grading. They present a FinQA demo, showcasing how Agent RFT improves performance and reduces tool calls, followed by customer success stories from Cognition, Ambience, Genspark, Mako, and Rogo, highlighting diverse applications and significant performance gains. The session concludes with advice on achieving success with Agent RFT and a Q&A.