Skip to content
Simon Place
FR EN
Start the courseStart
Book a Meet call

Building AI Agents

Three parts and ten modules for developers, each in five lessons: the essentials, going deeper, videos, practice and a check. One running project: a refund agent to build and measure.

Start the course Try an exercise

Free, no account needed.

async function runAgent({ model, tools, prompt }) {  const messages = [{ role: 'user', content: prompt }];  while (true) {    const response = await model.create({ messages });    messages.push({ role: 'assistant', content: response.content });    if (response.stop_reason === 'end_turn') {      return response.content.find((b) => b.type === 'text').text;    }
Repair the agentFix the tool loop
Open in the editor

The program

Part 1 of 3

The foundations

Understand what an agent really is, what it is made of and how it acts on the world.

3 modules. About 2 h 15.

  1. What an agent really isTell a model, a program and an agent apart, and know when an agent is the wrong choice.About 40 min
  2. Anatomy: components and memoryName the modules of an agent and choose the types of memory it actually needs.About 40 min
  3. Tools and MCPFollow a tool call from end to end and design tools a model uses correctly.About 55 min

Part 2 of 3

Building

Pick an architecture, split the work when it pays off and connect the agent to your data.

4 modules. About 3 h 45.

  1. Reasoning loops and architecturesChoose between ReAct and planning up front, then between one agent and a more complex setup.About 55 min
  2. Multi-agent systemsKnow when several agents are worth their coordination cost, and which structures are available.About 45 min
  3. Choosing a frameworkChoose by type of system rather than popularity, and know when to skip a framework.About 55 min
  4. Agentic RAGKnow what an agent adds to RAG, and when it only weighs the system down.About 1 h 10

Part 3 of 3

Reliability and delivery

Take an agent from a working demo to something you can trust, measure and ship.

3 modules. About 3 h 25.

  1. Governance and securityIdentify the risks an agent adds and put guardrails in the code, not just in the prompt.About 50 min
  2. Evaluating an agentBuild an evaluation set that tells you whether a change really improves the agent.About 1 h 45
  3. Use cases and final projectSpot real opportunities, avoid needless agents and ship a measured agent.About 50 min

Watch an agent work

The customer writes: “Order 4512 arrived broken. I want a refund.”

  1. Thought I’m missing the order details. I’ll look them up.

    Nobody wrote this path: the model chooses its next step. What an agent really is

  2. Action get_order(id=4512)

    The model asks, your code executes. Tools and MCP

  3. Observation { "amount": 89, "delivered": "3 days ago" }

    This result enters working memory, which is limited. Anatomy: components and memory

  4. Thought I don’t know the rule. I’ll check the policy before deciding.

    The agent replans after each observation: that’s the ReAct loop. Reasoning loops and architectures

  5. Action search_policy("damaged item")

    The agent picks the source to query by itself. Agentic RAG

  6. Observation "Full refund within 30 days."

    A framework can run this loop for you. Choosing a framework

  7. Thought This case qualifies. I can prepare the refund.

    A second agent could check this reasoning. Is it worth the cost? Multi-agent systems

  8. Action refund(order=4512, amount=89) // awaiting approval

    Irreversible action: the code requires a human, whatever the model says. Governance and security

  9. Observation "Approved by customer service. Refund issued."

    It all worked this time. How do you know it will across a thousand requests? Evaluating an agent

  10. Answer You’ve been refunded $89. It should arrive in 3 to 5 days.

    Did this task really need an agent? Use cases and final project