The Blueprint for Smart AI
Agentic AI excels at reasoning, planning, and tool use to tackle challenges. This graphic outlines key design patterns enabling these cutting-edge systems.
Anatomy of an AI Agent
An agent goes beyond being a mere model—it's a cohesive system of interworking parts. At its center, the Large Language Model (LLM) serves as the 'Planner,' driving reasoning and coordination among essential components.
Memory
Stores conversation history and learned knowledge for context.
Tools
Tools and features the agent can access, such as calculators or web search.
Profile
Instructions defining the agent's identity, goals, and constraints.
Core Design Patterns
Patterns are repeatable methods shaping an agent's thinking and behavior. They serve as core frameworks for designing effective agents.
ReAct
The agent verbalizes its thoughts prior to acting, ensuring its reasoning is clear.
Reflection
The agent reviews and refines its output, achieving superior results.
Multi-Agent
A team of specialized agents collaborates to solve complex problems.
How to Choose a Pattern
The ideal pattern varies by goal. This chart outlines key patterns and factors to guide you in choosing the best blueprint for your project.
ReAct Pattern In Action
Here's a brief overview of how an agent applies the ReAct pattern to solve a query, cycling through reasoning, action, and observation.
USER GOAL
"What's the weather in Paris?"
THOUGHT
I need to check the weather. My weather tool can help me.
ACTION
search_weather(city="Paris")
OBSERVATION
API returned: { "temp": "15°C", "condition": "Sunny" }
THOUGHT
I know the solution. I'll reply to the user now.
FINAL ANSWER
The weather in Paris is 15°C and sunny.