The agent alternates between generating a Thought (reasoning about what to do) and taking an Action (calling a tool). The Observation from the tool is fed back into context. This cycle repeats until the task is complete.
Flow Diagram
Task Input
THOUGHT
(reason about next step)
ACTION
(call tool)
OBSERVATION
(tool result)
Final Answer
When To Use
Best for tasks requiring dynamic tool selection and multi-hop reasoning - Q&A over multiple sources, code debugging, data lookup chains.
Pattern
Show / Hide
Quick Ref
ReAct:Interleaved reasoning + action
Plan-and-Execute:Upfront plan, then execute steps
Reflection:Self-critique and revise output
Tool-Use:Select and call external tools
Multi-Agent:Delegate subtasks to specialized agents
Agentic AI systems are built from a small set of composable design patterns. ReAct interleaves reasoning and actions. Plan-and-Execute generates a full plan upfront then executes steps. Reflection uses a critic LLM to improve outputs. Tool-Use grounds the agent in real-world data. Multi-Agent delegates subtasks to specialized sub-agents. Understanding when to apply each pattern is the difference between a brittle prototype and a production-grade agent.
Five patterns: ReAct, Plan-and-Execute, Reflection, Tool-Use, Multi-Agent - click any card to see its flow diagram
Each pattern includes a 4-6 node flow diagram showing the complete execution path
Toggle "when to use" descriptions to understand which tasks each pattern is optimized for
Toggle tradeoffs table to compare pros and cons - cost, adaptability, debugging difficulty
Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.