Build Your First AI Agents

Objectives

By the end of this course, you will be able to:

• Tell a conversation, a custom assistant and a connected agent apart, and say what each one does without you
• Write an agent’s mission sheet: what it does, from what, what it never does, when it hands over to a person
• Write structured instructions (role, context, sources, steps, format, rules, edge cases) and rules that never contradict each other
• Prepare reference documents an assistant reads correctly, and have it cite its source or answer “I can’t find this information” rather than make things up
• Build the same assistant in a ChatGPT project and in a Claude project from one instruction file, and place Gemini
• Design a test set (common cases, edge cases, out of scope, traps, sensitive data), run it before you trust the agent, and run it again after every change
• Build and test two complete agents (draft replies to customer emails, sorting supplier invoices), then a third on your own, guided, checked against the answer key
• Spot an invented answer (a rule, a price, a date) and correct it from the reference document
• Decide what you hand to an agent and what you never do, set memory, sharing and connectors, and keep each agent’s tracking sheet
• Say when an agent is ready to be connected to your tools, and what that changes

Course outline

An introduction lesson, then four modules: 21 lessons in all. Open a module to see its lessons.

1. Module 0Introduction lesson1 lesson · 8 min

   1. What goes in, what comes out: three agents, and what this course teaches you Free sample lesson8 min
2. Module 1The basics: what an agent is, its instructions, its documents and its tests6 lessons · 34 min

   1. From conversation to agent: four steps5 min
   2. The mission sheet: what the agent does, what it never does, when it hands over5 min
   3. Instructions in seven blocks: role, context, sources, steps, format, rules, edge cases7 min
   4. Rules that hold: say what to do, explain why, show an example, never contradict yourself5 min
   5. Reference documents: prepare them, have the source cited, or have it say “I can’t find this information”6 min
   6. Test before you trust: the test set, the grid, the second run5 min
3. Module 2Agent 1: answering your customers’ emails, in ChatGPT and in Claude7 lessons · 44 min

   1. Where to build an agent as of October 18, 2026: projects, Gems, GPTs and skills5 min
   2. Agent 1: the mission, the company’s file and the four-file kit6 min
   3. Agent 1: the instructions, block by block6 min
   4. First step in ChatGPT: the project, the instructions, the documents, memory, the first try7 min
   5. The same build in Claude; Gemini as a landmark6 min
   6. Agent 1: fifteen test emails, what fails on the first run, and why7 min
   7. Agent 1: correct, run again in both tools, use it every day6 min
4. Module 3Agent 2: filing your supplier invoices; the third agent as a guided exercise4 lessons · 25 min

   1. Agent 2: the mission and the company’s filing plan6 min
   2. Agent 2: the instructions, block by block5 min
   3. Agent 2: build and try it in Claude; reference documents and invoices to process7 min
   4. Agent 2: test on ten documents, correct, use it every week7 min
5. Module 4Stay in control, and what you never hand to an agent3 lessons · 16 min

   1. Stay in control: the tracking sheet, review, regular testing5 min
   2. What you never hand to an agent: data, decisions, commitments, access7 min
   3. What comes next: when to connect your agent to your tools4 min

How the course is made

The agents work on two fictional companies, an online tea shop and a renovation tradesman, with their documents, emails and invoices, all marked as fictional. You redo each build in your own account, then adapt an agent to your own work.

Building in ChatGPT and in Claude is shown on real, dated screenshots. The other lessons rely on the vendors’ official help pages, cited and dated. Each instruction file says at the top which tools it was tried in, and asks you to run the test set in your own tool before using it.

No number of hours saved is promised, nor an agent that would work without review: the agents in this course prepare, and you decide.

Who it is for

Anyone who already uses ChatGPT or Claude and wants an assistant that works for them: freelancers, employees, small teams, consultants. Having written requests to an assistant is enough: the course explains each term the first time it comes up. You need a computer, an up-to-date browser and a free ChatGPT or Claude account; paid features are flagged with their price. The course does not cover programming, agents that act on their own in your tools (email, calendar, spreadsheet), or an assistant installed on your website to answer your customers directly.