LLM workflows
In agent patterns, we explored the common AI agent patterns, each built around a set of modular capabilities: perception, action, learning, and cognition. At the heart of the cognitive module in many agent patterns is a large language model (LLM) that is capable of reasoning, planning, and decision-making. However, invoking an LLM alone is not sufficient to produce intelligent, goal-directed behavior.
To perform complex tasks reliably, agents must embed the LLM within a structured workflow, where the model's capabilities are augmented with tools, memory, planning loops, and coordination logic. These LLM workflows allow an agent to break down goals, route subtasks, call external services, reflect on results, and coordinate with other agents.
This chapter introduces the core design patterns for building robust, extensible, and intelligent LLM-driven cognitive modules, organized around reusable workflows.
In this section
