In this session, we explore how to make the latest generation of AI systems do more than answer questions: we put them to work as AI agents that independently carry out more complex tasks using code, documents, data and internal tools. We look under the hood of language models and agents to understand where their limitations come from and how to overcome them. We then dive deeper into how to steer agents more effectively with context and reusable skills, how to give them access to your own systems through connectors, and where things go wrong with permissions and data leaks. We apply all of this through live demos and exercises, so that everyone can use agents optimally and responsibly in their own work.