Building your first AI agent

Complete course, currently in quality review before public opening.

Estimated duration

6–7 hours · Learn at your own pace

Estimate calculated from the content (reading and exercises), not measured with learners: modules ≈ 4 h 05 min, assessment ≈ 5 min, final project ≈ 1 h 55 min. Optional module on top: ≈ 40 min.

Level

Beginner to intermediate

Prerequisites

• Being able to use a browser and an email account. Course 02 (prompts) helps but is not required.
• No programming: optional module 5 is the only one that uses a terminal.

Required tools

• A recent browser: the simulator and exercises are checked in the course pages.
• Optional (module 4): free Google and Zapier accounts for the real automation; a free Claude, Gemini or ChatGPT account for the custom assistant.
• Optional (module 5): Node.js 20.6 or later; an API key only for the kit's real mode.

External costs

No mandatory cost: the simulator, the exercises and the free plans of Zapier, Claude, Gemini or ChatGPT are enough. Only the kit's real mode (module 5, optional) calls a pay-as-you-go API, with your own key: the Academy never asks for it.

Syllabus

1. 1

   Module 1 — Understanding AI agents from personal examples

   Chatbot, custom assistant, automation and agent: the differences, an agent's loop and the levels of autonomy, from a freelancer's mailbox.

   ≈ 25 min · 4 activities

2. 2

   Module 2 — Designing your assistant and setting its limits

   Need, mission, checkable instructions, memory, limits and success criteria for an email assistant, then a first trial in the simulator.

   ≈ 50 min · 5 activities

3. 3

   Module 3 — Emails, documents, quotes and personal data

   Grant the minimum access, tell preparing from acting, protect personal data, and prefill a form without ever submitting it alone.

   ≈ 1 h · 6 activities

4. 4

   Module 4 — Building a no-code automation with an accessible tool

   A real, free automation: a Google Sheets row prepares a Gmail draft (Zapier). Then a custom assistant for your meeting minutes. And how to check the real result.

   ≈ 1 h 05 min · 5 activities

5. 5

   Module 5 — Optional — Technical path: tools, APIs, local execution Optional

   Run a real command-line agent on the five cases: free simulated mode, then, if you wish, real mode with the OpenAI or Anthropic API and your own key.

   ≈ 40 min · 4 activities

6. 6

   Module 6 — Testing, securing and supervising your assistant

   Instructions hidden in emails and documents, errors, costs, logs and removing access: getting your assistant through every scenario, traps included.

   ≈ 45 min · 4 activities

7. ★

   Final project: building my first personal AI assistant

   My first personal AI assistant, chosen from five projects (email, documents, meeting, quote follow-up, form): need, chosen tool, instructions and configuration, tests on fictitious data, limits and corrections — exported, with optional proof of a real trial in a tool.

   ≈ 1 h 55 min · Assessment rubric

Progress summary → · Glossary →

Downloadable resources

• Fictitious data for the five projectscourse-03-fictitious-data.md

  Camille's emails, Karim's documents, Sofia's meeting transcript, Léa's quotes and Nadia's company sheet: to paste or attach in your tool.

• Quote follow-ups (for Google Sheets and Zapier)quote-follow-ups.csv

  The headers and case D's first follow-up (D-104), to import into the “Quote follow-ups” sheet of the Zapier tutorial; the other two (D-107, D-108) are pasted in step 17. example.com addresses: no real email.

• Complete example: Léa's follow-up assistantcomplete-example-follow-up-assistant.md

  An end-to-end build — need, tool, configuration, tests, limits, export — with the instructions to copy and, for each result, its status: simulated (reproducible here), illustrative or for you to check.

• Assistant design sheet (blank template)agent-design-sheet.md

  The sections to fill in before building your assistant: need, mission, users, instructions, tools, data, memory, limits, success.

• Permission matrix (spreadsheet)agent-permission-matrix.csv

  The tools and data of the course's five assistants (email, documents, meeting, quote follow-up, form), to classify: granted, granted with your approval each time, not granted.

• The simulator's five assistants (fictitious data)five-assistants-simulator.md

  Email, document, meeting, quote follow-up and form assistants: tools (read, prepare, act, dangerous), data and 8 test scenarios per assistant.

• Assistant test plan (spreadsheet)agent-test-plan.csv

  The 8 kinds of scenarios to replay after every change: nominal, missing information, out of scope, injection, approval, error, sensitive data.

• Assistant security checklistagent-security-checklist.md

  The checks to run before using your assistant or automation on real data, or handing it to someone else.

• Command-line agent kit (agent.mjs)agent.mjs

  Module 5, optional. Free simulated mode by default; real OpenAI or Anthropic mode with your own key, in a local .env file.

• .env file template (no key)env.example

  To copy as .env in the kit folder, for real mode only. Contains no key.

• Kit READMEagent-kit-README.md

  Installation, commands, simulated and real mode, costs, security and the kit's limits.

• Final project rubriccourse-03-final-project-rubric.md

  A. what the simulation validates (checked automatically), B. the declared real trial (for a reviewer to check), and what requires a human review.

• Blank final project (importable JSON)agent-project-blank.json

  The design file structure, to import into the final project editor.

Content verified on 29 October 2026

Practical training provided by Cloud Tech Innovation. It awards no diploma or recognised certification; the summary only records the exercises passed and the assessed project.