Chapter 7: What Are Agentic Patterns?
Patterns are reusable solutions to common problems. In software engineering, design patterns give developers a shared vocabulary and proven approaches to recurring challenges. Agentic patterns do the same for AI systems.
Rather than inventing from scratch every time, you can draw on established patterns that others have tested and refined. This accelerates development, reduces risk, and makes your solutions easier to explain and maintain.
Patterns aren't rigid templates — they're starting points. You adapt them to your specific context, combining and modifying as needed.
Four Categories of Patterns
We'll cover four categories:
- Single-agent patterns — how one agent reasons and acts
- Multi-agent patterns — how multiple agents work together
- Human-in-the-loop patterns — how humans and agents collaborate
- Memory and context patterns — how agents access and retain information
The single-agent patterns in Chapter 8 apply to the baseline framework architecture described in Chapter 4. The multi-agent patterns in Chapter 9, and the shared memory patterns in Chapter 11, describe how that architecture expands when multiple specialist agents collaborate. If you are building a single AI advisor or assistant, Chapter 8 is your primary reference. If you are coordinating multiple agents across a complex workflow, Chapters 9 and 11 become equally important.
If you're building a single AI advisor or assistant — like the specialists in the Pragmatix Advisory Portal — the single-agent patterns are your primary toolkit.
Multi-agent patterns become relevant when you need to coordinate multiple specialists to handle complex, cross-domain work.