1 What are LLM Agents and Multi-Agent Systems?
This chapter covers
- Current real-world applications of LLM agents and multi-agent systems
- What LLM agents are and why LLMs alone are insufficient
- Important design patterns, enhancements, and protocols for LLM agents
- When applications may benefit from multi-agent systems
- A roadmap for developing LLM agents and multi-agent systems
If a user asks a Large Language Model (LLM) where to find the best value in croissants in New York City, the LLM might respond, “I will search the web for highly-rated croissants and their prices.” LLMs are very good at expressing intent to act toward a specific goal—to generate text that tells us what they are going to do to resolve a query. At this point, however, we run into a critical limitation: since LLMs are only text-generators, they cannot act on their intentions. They can articulate a plan for processing a task, but cannot carry it out—unless they are surrounded by a system to orchestrate the plan and execute the actions.
These orchestration systems are called LLM agents. We’ll add some nuance to this definition soon, but for our purposes, LLM agents are systems that automatically turn the LLM’s intentions into actions.