Working through this module?0 of 19 done
Module 14 · Page 15 of 20 · 9 sections

Agentic AI Design Patterns and Architectures

Reusable patterns: ReAct, plan-and-execute, reflection, multi-agent, tool use and human-in-the-loop.

05Planning and Reasoning in Agentic AI08Multi-Agent Systems in Agentic AI03Types of Agents
14.1

What are Agentic Design Patterns?

  • Design patterns are reusable solutions for common problems in building agentic AI systems.
  • They provide a structured way to design, build and scale agents.
  • These patterns help in better modularity, maintainability and reliability.
  • Useful for real-world applications and production systems.
14.2

Common Design Patterns

1
ReAct Pattern
Agent thinks, acts and observes the result.
2
Plan-and-Execute
Agent first creates a plan, then executes the steps.
3
Reflective Pattern
Agent reviews and improves its own output.
4
Multi-Agent Pattern
Multiple agents collaborate to solve complex tasks.
5
Tool-Use Pattern
Agent uses external tools (APIs, DB, search, etc.).
6
Human-in-the-Loop
Human provides feedback or approval in between.
7
Memory-Augmented
Agent uses memory for better context and personalization.
14.3

ReAct Design Pattern

  • ReAct = Reasoning + Acting.
  • Agent first thinks (reasoning), then takes an action using a tool, observes the result and continues.
  • Useful for dynamic problem solving and tool use.
User Query
ReasoningThink
ActionUse tool
ObservationResult
Update Reasoning and Continue
Final Answer
Related05Planning and Reasoning in Agentic AI
14.4

Plan-and-Execute Pattern

  • Agent first creates a high level plan to achieve the goal.
  • Then executes each step using tools or sub-agents.
  • Useful for complex tasks with multiple steps.
User Goal
Planner AgentBreak into steps
Task Queue
Execution AgentExecute each step
Final Output
Example plan
  1. 1Search best hotels in Goa
  2. 2Compare prices
  3. 3Check user reviews
  4. 4Prepare final recommendation
  5. 5Share with user
14.5

Reflection Pattern

  • Agent reviews its own output and identifies mistakes or areas of improvement.
  • Useful for improving accuracy and reliability.
  • Can use the LLM itself for self-critique and refinement.
Initial Answer
ReflectionReview and critique
Improved Answer
14.6

Multi-Agent Collaboration Pattern

  • Multiple agents work together, each with a specific role.
  • Agents can communicate and share information.
  • Useful for complex tasks that require different skills.
Coordinator AgentAssign tasks and manage flow
Research AgentSearch information
Analysis AgentAnalyze and process
Writer AgentGenerate final output
Final Result
Related08Multi-Agent Systems in Agentic AI
14.7

Tool-Use Pattern

  • Agent uses external tools like web search, database, APIs, code execution, etc.
  • Improves the capability of the agent beyond just the LLM.
  • Tools can be selected dynamically based on the user query.
User Query
AgentLLM + tool selection
Web Search
Database
API Call
Code Execution
Custom Tool
Final Answer
Related06Tool Use and Function Calling in Agentic AI
14.8

Human-in-the-Loop Pattern

  • Human provides feedback, approval or guidance during the agent workflow.
  • Useful for sensitive or critical tasks.
  • Helps in reducing errors and improving trust.
Agent Output
Human ReviewApprove / modify
Updated Output
Related12Security and Safety in Agentic AI
14.9

Key Takeaways

  1. 1Design patterns help in building robust and scalable agents.
  2. 2Choose the right pattern based on the problem and use case.
  3. 3Combine multiple patterns for better results.
  4. 4Use tools, memory and human feedback for real-world systems.
  5. 5These patterns are useful for enterprise and production grade agentic AI applications.