Part 2: Patterns
Chapter 12: Choosing the Right Patterns
Patterns aren't one-size-fits-all. Choosing the right patterns depends on your specific context:
| Factor | Consideration | Pattern Guidance |
|---|---|---|
| Task complexity | How many steps? How much reasoning? | Simple tasks → single-agent patterns Complex tasks → multi-agent or pipeline patterns |
| Risk level | What's the cost of errors? | High risk → human-in-the-loop patterns Low risk → more autonomy |
| Domain breadth | How many areas of expertise? | Narrow domain → single specialist Broad domain → multiple specialists with routing |
| Quality requirements | How important is output quality? | High quality → reflection, consensus patterns Speed priority → simpler patterns |
| Data needs | What information does the agent need? | Static knowledge → baked into DNA Dynamic/proprietary data → RAG patterns |
The Pragmatix Approach
Start simple and add complexity only as needed:
- Begin with a single agent using ReAct and tool use
- Add reflection if quality needs improvement
- Introduce human-in-the-loop patterns for high-risk decisions
- Scale to multi-agent only when a single agent's DNA becomes unmanageable
Key Principle
Complexity has costs — in development, maintenance, and debugging. Only add complexity when simpler approaches have proven insufficient.