When to Use Each Design Pattern
Match your use case characteristics to the right pattern before building.
Pattern Selection by Use Case
Most enterprise agent implementations can be matched to one or two primary patterns based on the characteristics of the use case. Use this table as a starting point, then read the individual pattern pages for implementation guidance.
| Characteristic | Primary Pattern | Secondary Pattern |
|---|---|---|
| Sequential steps, dependent on each step's result | ReAct | Chain of Thought |
| Long-horizon task with a plannable sequence | Plan-and-Execute | Human-in-the-Loop |
| High-stakes output that must be verified | Reflection | Human-in-the-Loop |
| Multiple independent data lookups needed | Parallelization | ReAct |
| Task spans multiple specialist domains | Multi-Agent Collaboration | Orchestrator-Subagent |
| Single entry point for diverse request types | Router | Multi-Agent Collaboration |
| Agent needs context from past sessions | Memory-Augmented | ReAct |
| Complex tool selection with many options | Tool Use | Router |
| Clear hierarchical task decomposition | Orchestrator-Subagent | Plan-and-Execute |
| Irreversible or high-value actions | Human-in-the-Loop | Reflection |
| Complex reasoning with audit requirements | Chain of Thought | Reflection |
The Most Common Pattern Combination
The most common production enterprise agent combines three patterns: ReAct (the core reasoning loop), Tool Use (the MCP tool set design), and Human-in-the-Loop (approval gates for high-stakes actions). This combination covers the majority of enterprise use cases and is the right starting point for most implementations.