Modern AI Engineering

Module 11 · Build

Autonomous AI Agents

In this module, we will learn how an LLM goes from answering questions to actually doing work. We will start with a single agent, see how it uses tools and memory, and then move to systems where many agents work together.

By the end of this module, we will be able to design a single agent, give it tools and memory, and scale it to a multi-agent system.

Lessons

  1. 11.1 AI Agents: Autonomous Decision-Making Systems: The Big Picture · What is an AI Agent · AI Agent vs Plain LLM vs Chatbot · The Five Core Parts · How an AI Agent Works End to End · A Concrete Example: Research Agent · Types of AI Agents · What AI Agents Can Do Today · When to Use an AI Agent · Common Failure Modes · Quick Summary
  2. 11.2 Function Calling: Giving LLMs Tools to Act on the World: What is Function Calling · Why We Need Function Calling · The Key Insight: The Model Does Not Run the Function · How Function Calling Works Step by Step · A Concrete Example: get_weather(city) · The Conversation Loop · Multi-Step and Parallel Function Calling · Relation to Structured Outputs and JSON Mode · Real-World Use: The Backbone of AI Agents · Quick Summary
  3. 11.3 The Agent Loop: Observe, Think, Act, Repeat: The Big Picture · What is the AI Agent Loop · Why an AI Agent Needs a Loop · The Think-Act-Observe Cycle · The Loop Step by Step · The Loop in Real Code · Parallel Tool Calls in One Turn · How the Loop Knows When to Stop · Common Loop Failures · Quick Summary
  4. 11.4 ReAct Agents: Interleaving Reasoning and Acting: What is a ReAct Agent · ReAct Agent vs AI Agent · Anatomy of a ReAct Agent · The ReAct Prompt Template · How a ReAct Agent Thinks and Acts · A Full Trace Example · Implementing a ReAct Agent · Common Failure Modes and How to Fix Them · Quick Summary
  5. 11.5 Plan-and-Execute: Tackling Complex Tasks in Two Phases: What is a Plan-and-Execute Agent · Plan-and-Execute Agent vs AI Agent · Anatomy of a Plan-and-Execute Agent · How a Plan-and-Execute Agent Works · A Full Trace Example · Plan-and-Execute Agent vs ReAct Agent · Common Failure Modes and How to Fix Them · Quick Summary
  6. 11.6 Reflection Agents: Self-Critique for Higher-Quality Outputs: What is a Reflection Agent · Reflection Agent vs AI Agent · Anatomy of a Reflection Agent · How a Reflection Agent Works · A Full Trace Example · Reflection Agent vs ReAct Agent · Common Failure Modes and How to Fix Them · Quick Summary
  7. 11.7 Agent Memory: Short-Term, Long-Term, and Episodic: The Big Picture · Why AI Agents Need Memory · The Memory Stack · The Four Core Operations · How Memory Flows at Runtime · What to Store and What Not to Store · Common Mistakes and How to Fix Them · Quick Summary
  8. 11.8 Model Context Protocol: A Standard Interface for Agent Tools: The problem before MCP · What is MCP? · MCP = Model + Context + Protocol · The USB-C analogy · How it is different from a normal API · The three parts of MCP · How it all works step by step · How the connection happens · A real example · Importance of MCP · Things we must be careful about · Summary
  9. 11.9 Agent Skills: Reusable Capabilities in Agentic Systems: The problem before Agent Skills · What are Agent Skills? · What is inside a Skill? · The description is the trigger · Progressive disclosure, the main idea · A Skill can carry real code · Where Skills live · How do we create our own Skill? · Agent Skills vs MCP · A real example · Importance of Agent Skills · Things we must be careful about · Summary
  10. 11.10 Open Knowledge Format: Structured Agent-to-Agent Communication: The problem: our knowledge is scattered · What is OKF? · OKF = Open + Knowledge + Format · What is inside an OKF bundle? · The frontmatter and the one required field · Cross-links turn files into a graph · Why plain markdown files? · How an agent actually uses it · OKF, MCP, and Agent Skills · What ships with OKF today · Summary
  11. 11.11 Multi-Agent Systems: Dividing Work Among Specialist Agents: The Big Picture · What is a Multi-Agent System · The Three Pillars · Common Agent Roles · How Agents Communicate · How Agents Coordinate · Multi-Agent vs Single Agent - The Trade-offs · Common Mistakes · When to Use a Multi-Agent System · Quick Summary
  12. 11.12 Subagents: Delegating Tasks Within an Agent Network: What is an AI Agent? · What are AI SubAgents? · Why do we need SubAgents? · How do SubAgents work? · Example use case · Benefits of using SubAgents · Challenges with SubAgents · Best practices
  13. 11.13 Agent Communication: Protocols and Message Formats: What is agent communication? · Why do agents need to communicate? · What agents need in order to communicate · How a message flows between agents · The ways AI agents communicate · Direct Communication · Centralized Communication · Broadcast Communication · Shared Memory Communication · What a message looks like · The rules agents follow to talk · Challenges when agents communicate · Best Practices
  14. 11.14 AI Orchestration: Coordinating Agents, Tools, and Flows: What is AI Orchestration? · Why do we need AI Orchestration? · AI Orchestration vs AI Agents · Components of AI Orchestration · How AI Orchestration works · Patterns of AI Orchestration · Sequential Pattern · Parallel Pattern · Conditional Pattern · Loop Pattern · Orchestrator-Worker Pattern · Tools for AI Orchestration · Challenges in AI Orchestration · Best Practices
  15. 11.15 Sakana Fugu: Lessons from an Open-Source Agent Study: What is Sakana Fugu? · Why Fugu was needed · The big picture: what Fugu does · Collective Intelligence · The two Fugus - Fugu and Fugu-Ultra · How Fugu picks the right model - the lightweight selection head · Teaching Fugu who is best - supervised fine-tuning · Polishing Fugu on real tasks - evolutionary strategies · How Fugu-Ultra conducts an orchestra - the Conductor · Teaching Fugu-Ultra to conduct - GRPO · Stopping the agents from copying each other · How well does Fugu perform? · The clever strategies Fugu discovered on its own · Quick Summary
  16. 11.16 Computer-Use Agents: Controlling Interfaces with AI: What is a computer-use agent? · Why do we need a computer-use agent? · The perceive, think, act loop · How does the agent see the screen? · How does the agent decide what to do? · How does the agent take actions? · A step-by-step walkthrough with an example · The system prompt and tools · Safety and guardrails · Limitations of computer-use agents · Conclusion

← Module 10: Building RAG Systems · Module 12: Agent Patterns and Frameworks →