Practical MLflow for Generative AI on Databricks

Chapter 7. Advanced Agents and Tools

In the previous chapters, we worked with a simple tool-calling agent that queries Databricks Vector Search to answer FAQ questions about Unity Airways. Agents, often referred to as AI agents, represent a major shift in GenAI application development. They can plan, invoke tools, observe results, and iterate based on feedback, enabling autonomous decision making and multistep reasoning. Because agents can adapt to complex scenarios without relying on predefined workflows, they can reduce tasks that once required time-consuming human intervention to just seconds. Customer service centers, for example, stand to benefit significantly when resolving traveler requests such as flight cancellations or specific flight changes.

The rapid adoption of agents has made it one of the most discussed topics in the AI community, with major companies and research institutions investing heavily in agent frameworks and tooling development such as Model Context Protocol (MCP). This push for better frameworks has helped MLflow 3.x grow faster thanks to active demand and contributors, as developers are looking for solid tools to manage and deploy these more complex GenAI applications.

This chapter delves further into agents to help our Unity Airways assistant perform more complex tasks for our users. We’ll look at the fundamentals of agent development, how to use LangGraph with MLflow and switch models from different providers, as well as how to create tools and integrate ...