Agent Architectures

ReAct (Reason + Act)

Simple, reliable, widely used. The agent alternates between thinking and acting.

Pros: Simple, interpretable

Cons: Can loop, no explicit planning

Plan-and-Execute

The agent creates a plan first, then executes each step.

Pros: Better for complex tasks

Cons: Plans can be wrong

Reflexion

The agent does a task, critiques its own result, and retries.

Pros: Self-improving

Cons: Uses more tokens

Multi-Agent

Multiple specialized agents collaborate via an orchestrator.

Pros: Specialization, parallelism

Cons: Complex to build

Choosing the Right Architecture

Your Turn!

Which architecture for each?

Customer support with order database

Generating a 50-page technical report

Debugging complex code