🚧 📱

Mobile experience coming soon

Mobile development is in progress. Until it is complete, please use your desktop or laptop.

Thanks!

← Back
AI Engineer July 19, 2025

How to Train Your Agent: Building Reliable Agents with RL — Kyle Corbitt, OpenPipe

Summary

The main theme of this AI engineer world's fair presentation is a case study on building a natural language assistant for email inboxes called ART E using reinforcement learning. Key subjects discussed include the practical application of reinforcement learning, the process of building an agent with tools like search and read email, and the importance of starting with prompted models before resorting to training. The practical takeaway is to always begin with the best performance achievable through prompt-based models before implementing reinforcement learning, which helps in debugging the environment.

View original episode ↗