The course
Build Compass yourself.
Compass is where people and a team of AI agents work together. In 41 short lessons you build it on Copilotz, one part at a time, and learn how every part works.
Your teacher is North, an AI agent, and the first one you'll build.
The course
US$ 29one payment
- 41 video lessons, over two hours in all
- In English and in Portuguese
- The code at the end of every lesson
- For developers who know TypeScript
Free lesson
Watch the first lesson.
Lesson 1.1, One agent, one message. North introduces the course, you send your first message to an agent, and you read the record of every step it took.
Why a course
Tutorials teach the pieces. Here you build the whole.
An app with agents is many pieces at once. Most tutorials teach one each and leave the fitting to you. In the course you build one app, and each piece arrives when the app needs it, with the reason for it.
- A model
- Tools
- Memory
- Search by meaning
- A database
- Documents
- A board
- Schedules
- The web
- Sign-in
- Tests
- Logs
CompassOne app, every piece in its place
What you build
A workspace for you and four agents.
North leads and asks the others for help: East builds to find out, South checks it against reality, West untangles hard decisions. They share a board, documents and a memory with you, in spaces you can share with people.
CompassSpaces
Personal
Launch team
Agents
Launch plan
Launch team
You: Can we launch on Friday?
North
Asking East for a draft and South to check the numbers.
South
The pricing page says 12 a month; the plan says 10. Which is right?
North
Holding Friday until you confirm the price.
Waiting for you
A board
Cards that you and the agents move.
Documents
With versions, and saves that merge or refuse.
Memory
Found by meaning, remembered and forgotten on purpose.
Skills and routines
Know-how on demand, and work on a schedule.
The web
Search, and pages read safely.
A computer
A terminal and files, with your permission.
In the last module you swap in Copilotz's own plugins and add sign-in, people, organizations and billing.
What you learn
Learn how a team works, and you can build one.
How agents work
What an agent sees on every turn, the tools it may use, what it remembers and forgets, and how four agents split the work.
One app, for people and for AI
The same board on the web, in a terminal, over REST and to other AI assistants over MCP, with the same rules for everyone: who may do what, and for whom.
Ready for real use
Tests on a fake model that need no key, PostgreSQL, a record of every step, sign-in and billing.
The lessons
13 modules, 41 lessons, one app.
Every lesson adds one thing to the app and ends with it working.
Part 1An agent and its app
Module 1 · 3 lessons
Hello agent
- 1.1One agent, one message
- 1.2North, in your terminal
- 1.3Compass in the browser
Module 2 · 5 lessons
Your own plugins
- 2.1Today, every turn
- 2.2One store
- 2.3Chats that name themselves
- 2.4Spaces
- 2.5Spaces on the web
Module 3 · 3 lessons
The board
- 3.1The board's data and Actions
- 3.2The board on screen
- 3.3The board for North
Part 2A workspace and a team
Module 4 · 4 lessons
Documents
- 4.1Documents and their versions
- 4.2Saves that merge, or refuse
- 4.3Documents on the web and in the terminal
- 4.4Documents for North
Module 5 · 3 lessons
Long conversations
- 5.1A line for every message
- 5.2The outline in North's prompt
- 5.3Zoom
Module 6 · 4 lessons
Memory
- 6.1Memories found by meaning
- 6.2Recall in the prompt
- 6.3Remember, recall, forget
- 6.4Memories on their own
Module 7 · 3 lessons
The team
- 7.1East, South and West
- 7.2Asking in parallel
- 7.3Reading other conversations
Module 8 · 3 lessons
Skills and routines
- 8.1Skills
- 8.2Routines on a schedule
- 8.3Agents that schedule
Module 9 · 2 lessons
The web for agents
- 9.1Search, an adapter port
- 9.2Reading a page safely
Part 3Out in the world
Module 10 · 2 lessons
Testing
- 10.1Fakes at the edges
- 10.2Parity
Module 11 · 3 lessons
Other ways in
- 11.1The REST API
- 11.2The client
- 11.3Your board over MCP
Module 12 · 3 lessons
Shipping
- 12.1PostgreSQL and blobs
- 12.2Watching it run
- 12.3A computer for your agents
Module 13 · 3 lessons
From one person to many
- 13.1Swapping in Copilotz's plugins
- 13.2Sign-in and people
- 13.3Many orgs and billing
How it works
Watch a few minutes. Write the code. See it work.
Every lesson, in six steps
- The pain
- The problem
- The solution
- Check it works
- What this unlocks
- Challenge
A short video
About three minutes each, with the real code and the app running.
The code at every step
Every lesson is a commit in the course's repository: start where the last one ended, and compare your code with ours.
Tests from day one
A fake model runs every test offline, so you can check your work without an API key. Add your own OpenAI key when you want North on a real model.
North is an AI agent, with an AI-generated voice.
Why now
Companies are hiring people who can do this.
On 7 October 2026 we read 34 job openings in the US, Europe and Brazil for people who put agents to work inside products. A few of them:
Pallet
Software Engineer Agent Delivery
Delivers agents for supply chain operations, from client discovery to integration, evaluation, and deployment. Uses Node, TypeScript, PostgreSQL, GCP, and web interfaces.
On-site in San Francisco or New York, five days a week; approximately 25% travel
See the listingcommercetools
AI Engineer
Builds agent loops, tools, structured outputs, business rules, failure handling, and MCP integrations with APIs. Includes generative UI, streaming, retrieval, and evaluation, with Python, TypeScript, and React.
Berlin, Munich, London, or Valencia; hybrid with three office days
See the listingSeso
Software Engineer AI Agents
Builds agents as part of a payroll and agricultural workforce management product. Includes end-to-end development, evaluation, and reliability.
Remote in the US, with a preference for San Francisco; application form asks about US residency
See the listingDecathlon Digital
AI Engineer
Deploys agents connected to corporate data and legacy systems within the supply chain team. Includes asynchronous execution, resilience, model routing, access restrictions, and cost management, with traces and evaluation.
Paris; listing mentions office options and two remote-work days per week
See the listingVestmark
Software Engineer
Develops workflows with agents that plan, execute, and iterate. The listing explicitly covers state, observability, testing of nondeterministic behavior, and inference costs, as well as the use of coding assistants.
Boston, hybrid, with at least three days a week in the office
See the listingWPP
Junior Engineer FDE Agentic and Eco System
Builds media-buying applications with agents, interfaces, pipelines, and integrations, from discovery to validation. The scope is described as production-lite, with observability, testing, and failure handling; it is not equivalent to taking ownership of mature, large-scale operations.
London, hybrid, with around four days in the office and international travel
See the listing
+64%
AI engineer hiring in the US in the second quarter of 2026, over a year before
LinkedIn Economic Graph, August 2026+178%
Freelance earnings from AI integration work, 2025 over 2024
Upwork, In-Demand Skills 2026
Public listings and reports; they don't endorse the course. Listings change.
Questions
Before you start
Who is it for?
Developers who build web apps and know TypeScript, APIs and a database. You don't need any experience with AI.
Do I need to know how to train models?
No. You use existing models inside an application: what they see, the tools they call, what they remember.
How is it different from the docs?
The docs are free, and they include this same app as a written guide, Build Compass. The course is that path on video: 41 short lessons taught by North, the code at the end of every lesson, and a challenge in each. The docs also have a shorter Getting Started that builds a booking app in 10 chapters.
Will I pay for models?
Not to follow along: every test runs on a fake model. To put North on a real model you add your own OpenAI key, and OpenAI bills you for what you use.
Will it help my career?
You finish with an application you can show and explain: agents, tools, memory and data working together, with tests. That's what the openings above ask for.
Are there free ways to learn this?
Yes, and good ones: Hugging Face's Agents Course, LangChain Academy, Microsoft's AI Agents for Beginners, and our docs. The course is for when you'd rather follow one path, on one app, with every decision explained.
When does it start?
Leave your email below and we'll send you the link to enroll as soon as enrollment opens. One payment of US$ 29.
The course
Reserve your seat in the course
Leave your email and we'll send you the link to enroll as soon as enrollment opens.
- 41 video lessons that build Compass
- In English and in Portuguese
- The code at the end of every lesson
US$ 29one payment