1 Understanding agentic applications
This chapter covers
- Introduction to agentic AI
- Augmented LLM function & tool calling
- CrewAI fundamentals
- Agentic AI production challenges
As a developer, you've probably interacted with Large Language Models (LLMs) - the AI models behind products like ChatGPT and Claude. On their own, these models do one thing: given a prompt, they predict and generate text. They can't research a topic across ten different sources, decide partway through that they need a different approach, or loop back when the results don't look right. Popular GenAI models can do some of these things, but only because their developers wrapped the underlying LLM in additional tooling and logic to make that possible. The raw model, by itself, generates text and stops.
Agentic AI, AI systems that augment LLMs with tools and autonomous reasoning, is about building that wrapping layer. In an agentic application, the LLM pursues goals autonomously. It reasons about what to do next, calls tools (like a web search or a database query), observes the results, and repeats until the task is complete. The LLM operates in a loop rather than a single exchange, and it retains control over how the work gets done.