Agent Architectures
AGAI 202
Survey the major architectural patterns for building AI agents. From simple ReAct loops to structured planning systems, learn how different architectures trade off capability, reliability, and interpretability.
How Agents Are Built
An AI agent is more than a language model with a prompt. It is a system with a specific architecture: a way of structuring perception, reasoning, memory, action, and feedback. Different architectures make different trade-offs, and choosing the right one for a given problem is a core engineering skill.
The Major Patterns
This course surveys the major architectural patterns in agentic AI: the ReAct loop, plan-and-execute architectures, reflection and self-critique systems, structured output workflows, and more. For each pattern, you will understand what problem it solves, where it shines, and where it breaks down.
What You Will Learn
You will implement the ReAct pattern from scratch, compare reactive and plan-and-execute architectures, and understand how orchestrator-subagent patterns enable complex task decomposition. You will learn how memory architecture choices affect agent behavior, how reflection and self-critique loops improve output quality, and how to evaluate architectural trade-offs for real use cases.
Who This Course Is For
This course is for engineers and architects who are building or evaluating AI agent systems and want to make informed design decisions — not just copy a pattern from a tutorial. A background in software development and basic familiarity with LLM APIs is assumed. This course is the foundation for everything that follows in the advanced and applied curriculum.
What you will learn
- Describe the ReAct pattern and implement it from scratch
- Compare plan-and-execute vs. reactive architectures
- Explain how reflection improves agent reliability
- Design a memory architecture appropriate for a given task
- Apply orchestrator-subagent patterns to complex tasks
- Evaluate architectural trade-offs for a real use case
Major topics
Why this course matters
Architectural decisions determine whether an agent is reliable, interpretable, and maintainable. Understanding the design space helps you make better engineering choices and debug failures more effectively.
Course modules
Foundational Agent Patterns
Begin with the core architectural patterns used by modern AI agents: ReAct loops and plan-and-execute systems. These patterns explain how agents alternate between reasoning, tool use, observation, and replanning.
Advanced Agent Patterns
Explore patterns that improve agent reliability and scale: reflection loops, orchestrator-subagent systems, and structured workflow agents. These architectures add control, specialization, and review to the basic agent loop.
Memory, Parallelism, and Architectural Evaluation
Complete the architectural toolkit with memory systems, task decomposition, parallel execution, and evaluation methods. Learn how to choose an architecture based on reliability, latency, cost, and risk.
Common misconceptions
There is one correct architecture for all agents
More complex architectures are always better
The ReAct pattern eliminates hallucinations
Architectures are fixed once the agent is deployed
Ask your AI guide
Ask anything about Agent Architectures, or choose a suggested question below.
AI responses are educational and may not be perfectly accurate. Press Enter to send, Shift+Enter for new line.
Related courses
Tool Use & Function Calling
Learn how AI agents extend their capabilities through tools. Explore function calling APIs, tool design patterns, and how agents decide which tools to use and when — turning language models into systems that can act on the real world.
Multi-Agent Systems
Explore the design and behavior of systems with multiple collaborating AI agents. Learn how agents communicate, coordinate, divide labor, and resolve conflicts — and how emergent behaviors arise when many agents interact.
Memory & Context Management
Learn how AI agents store, retrieve, and manage information across interactions. Explore the different types of agent memory — in-context, episodic, semantic, and procedural — and the techniques used to give agents effective long-term recall.