The Lab the Other Courses Need
There’s no shortage of good, free AI courses. What most of them don’t have is a lab bench: a real system you can break, check and play against, with whatever AI you bring. That’s the gap this one fills.
Kevin went shopping this week. Not for poker chips: for AI courses. He was about to offer this one to New Mexico schools, and he wanted to know what it was up against. The honest answer surprised him. It isn’t up against anything. The big courses are lectures, and this is a lab.
What the industry brings
First, credit where it’s due. The free courses out there are good, and some of them are huge.
| Course | Who it’s for | What it’s great at |
|---|---|---|
| Hugging Face Agents Course | People with basic Python | How agents think, act and use tools, plus the popular frameworks. The certificate is free, and you earn it. |
| Microsoft’s AI Agents for Beginners | Beginners who like reading code | Short lessons on design patterns, trustworthy agents and production, each with runnable Python. |
| Google and Kaggle’s 5-Day AI Agents Intensive | Anyone with a week and some Python | Scale and energy: more than a million people signed up in 2025, with live sessions, codelabs and a capstone. |
| Claude Academy | People using Claude | Free courses on Claude Code, MCP and agent skills, from the people who make it. |
| Harvard’s CS50 AI | Programmers with seven weeks | The classic ideas underneath: search, optimization, learning from rewards. |
| Elements of AI | Everyone, no code | What AI is and isn’t, in plain language. |
If you want to learn AI, start with any of them. Really.
What they mostly share
Look closely and most of them run on the same happy path. You open a notebook in someone’s cloud. You install their framework. You build a small agent that does what the lesson says, and the lesson ends when it works.
That isn’t a criticism. It’s how you teach a hundred thousand people at once. But it leaves out the part that matters most once you have a job: what happens when the AI is wrong, the setup breaks, or something is quietly cheating and every dashboard says green.
A chemistry class has a lecture and a lab. You need both. Most AI courses are all lecture.
Why this lab works
It doesn’t care which AI you bring
The Dealer (Part 15) only sees moves. Fold, call, raise. It has no idea what made the decision, and it doesn’t ask. A ten-line rules bot, a student’s Python script, an agent built on Claude, an agent built on some other model: they all sit at the same table through the same open API, and they all get the same cards.
That makes it AI agnostic in a way most courses can’t be. Kevin’s own AI player happens to run on Claude, on Amazon Bedrock, but nothing about the table depends on that. Swap in another model and the table won’t notice. The security lessons in Part 21 cover Microsoft and AWS side by side, because that’s the world students will actually work in. And the tools are open too: MCP for the agent’s connection to the Dealer, VS Code for the editing, plain text files for student bots.
Kevin wrote this with Claude. It isn’t a course about Claude. Bring whatever you’ve got.
It’s built to fail, on purpose
Most courses show you an agent that works. This one starts with bots that cheated for twenty years (Part 1), an AI whose first answer you can’t trust (Part 5), and a student table with three doors left open on purpose (Part 22). You learn how AI goes wrong by watching it go wrong, then fixing it.
It shows its receipts
Every claim ties to a numbered game. Look it up in Game Review, download the data, and check for yourself. In Part 20 the cheat is caught from the hand histories alone. You don’t have to take anyone’s word for anything, including Kevin’s.
It’s real, and it’s old
This isn’t a toy built for a lesson. It’s a poker engine from 2005, still running, with 18,689 files on the server and a disk that filled up this week (Part 24). Most jobs mean inheriting something old and making it better. Here you get to practice that, with an AI as your partner.
It runs without the internet
The whole table fits on a $140 Raspberry Pi kit and runs with the Wi-Fi off (Part 23). No accounts, no cloud bill, no API keys. That matters for a rural school, a kid at home, or a classroom where the network is locked down.
It’s a game
Lessons end. Games don’t. Six student bots in one file, a round robin where everyone gets the same cards, and a winner at the end. A club can come back every week with a better bot, and one school can play another. Nobody plays for money, ever.
It’s free, and it stays free
It’s licensed for noncommercial use: schools, clubs and anyone learning can use it, change it and share it. Nobody can sell it.
What it doesn’t do
To be fair, the lab has gaps, and the big courses fill them.
- No certificate. Hugging Face gives you one. Here, a teacher can sign one for winning the tournament.
- No videos, no Discord. It’s articles, a table and a Pi.
- One author. Kevin has a day job. That’s why it’s built to be handed off.
- No frameworks, no RAG, no neural networks from scratch. Go learn those from the people who built them.
That’s the point. You don’t have to choose. Take the lecture from them and the lab from here.
How to pair them
| If you are… | Take this first | Then come to the table for |
|---|---|---|
| New to AI, no code | Elements of AI | Playing the AI agent, reading its reasons, catching the cheat from the data |
| A beginner programmer | Microsoft’s AI Agents for Beginners | Writing your first bot and entering the student table |
| Comfortable with Python | Hugging Face, or the Kaggle intensive | Building an agent that plays a real opponent, then hacking the doors |
| Curious about the algorithms | CS50 AI | Opponent memory and small samples (Part 11); a bot that learns from its own wins and losses is still to come |
| Using Claude to build | Claude Academy | Running a whole project with an AI partner, the way Part 19 and Part 24 describe |
| A teacher | Any of the above, for you | A Pi kit, the spoiler folder, and a tournament to end the term |
The big courses teach you what an agent is. The table teaches you what to do when it lies to you.
For teachers and clubs
The lab is free for schools, clubs and anyone learning. Kevin will help you get started, mentor whoever teaches it, and give away the Pi release. Write to him at [email protected].
Sources (checked Oct 24, 2026; course details change, so check each course’s own page):
Written by Kevin Swinson with Claudette, the pen name for Claude, the AI from Anthropic that helped build HoldemRobots.AI. Describes the course landscape as of October 24, 2026. The other courses named here are not connected with HoldemRobots.AI.