Paid AI bootcamps can cost hundreds of dollars or more, yet some of the best material available is free, published by the companies and universities that do the work. The hard part is knowing what to take, and in what order, because a beginner and a working developer need very different things.
These six resources cover a full path. You can start without knowing how to code and finish by building and testing your own agents. I've put them in the order I'd take them, and I've included a realistic idea of who each one suits, so you can skip what you already know.
Costs change, so confirm the details on each page before you plan around them. The course content is free as listed here, but some extras are not, such as paid certificates, and hands-on work with AI models can need an API key that charges for usage.
Every entry below was found through the @RoundtableSpace feed on X, then checked against the project's own page on 4 November 2026. Numbers quoted are the projects' own claims. Read each license and README before you depend on anything here.
| Project | Best for | License |
|---|---|---|
| CS50P: Introduction to Programming with Python | Learning to code from zero | Free via OpenCourseWare |
| AI for Beginners | Understanding how AI works | MIT |
| Prompt Engineering Interactive Tutorial | Writing prompts that work | See repo |
| Introduction to Claude Cowork | Using AI on real files and projects | Free course |
| DeepTutor | A tutor for your own study material | Apache 2.0 |
| AI Engineering from Scratch | Building real AI systems | MIT |
1. CS50P: Introduction to Programming with Python
CS50P is Harvard's introduction to programming with Python. Its ten weeks cover functions and variables, conditionals, loops, exceptions, libraries, unit tests, file I/O, regular expressions, object-oriented programming, and a final week of everything else.
You can take it for free through OpenCourseWare by working through the material. Submitting problem sets and the final project for feedback requires an edX account, and verified or professional certificates go through edX.
Why start here in a list about AI? Because you can't review code you can't read. People who skip this stage end up stuck the moment an agent's output breaks, unable to tell a real fix from a confident guess. Python is also the main language of AI tooling, so nothing you learn here is wasted. Unit tests in week 5 are especially useful. Being able to write a test is the single best way to check what an AI gives you.
Do the problem sets by hand, without an assistant, at least for the first few weeks. It's slower, and that's the point. The struggle of fixing your own bug is what builds the skill that later lets you supervise an agent. If you already write code in another language, you can move through it quickly, or skip it.
2. AI for Beginners
Microsoft's AI for Beginners is a 12-week, 24-lesson curriculum with practical lessons, quizzes and labs. It's beginner friendly, and it covers tools such as TensorFlow and PyTorch, along with ethics in AI. The lessons run from the history of AI and knowledge representation through the perceptron, building your own small framework, overfitting, computer vision and beyond. Translations are available in many languages.
This course teaches concepts more than current products, which makes it age well. Once you've built a tiny neural network yourself, words like training, loss and overfitting stop being mysterious, and a lot of marketing stops working on you.
Some labs use notebooks, so you'll want basic Python first. You can work through them on a laptop, and the README explains setup in its first lesson. Take a lesson a week if you have a job, because the lessons need time to settle.
Not everybody needs all of this. If your goal is using AI tools at work, the early lessons and the ethics material are enough. If you want to be an AI engineer, do the whole thing, including the labs. It's a good base before you touch anything about agents, since it explains why models behave the way they do.
3. Prompt Engineering Interactive Tutorial
anthropics/prompt-eng-interactive-tutorial
Anthropic's interactive tutorial is a hands-on course in nine chapters, each with a lesson and exercises, plus an appendix. Beginner chapters cover basic prompt structure, being clear and direct, and assigning roles. Intermediate chapters cover separating data from instructions, formatting output, thinking step by step and using examples. Advanced chapters cover avoiding hallucinations and building complex prompts for chatbots, legal, financial and coding work. The appendix adds chaining prompts, tool use and search and retrieval.
The format is the reason to pick it. Every lesson ends with an Example Playground where you change the prompt and see what happens. You learn much faster by breaking a prompt than by reading about one. A version on Google Sheets exists too, if you'd rather not touch a notebook.
The tutorial uses Claude 3 Haiku in its examples, so you'll need an API key and a small amount of credit. Some model advice has aged, but the ideas in it hold up, particularly separating your data from your instructions and showing examples. Those two habits alone fix a large share of bad prompts. They also help with security, since keeping data apart from instructions is the first step in defending against prompt injection.
4. Introduction to Claude Cowork
anthropic.skilljar.com/introduction-to-claude-cowork
This free course from Anthropic is about working alongside Claude on your real files and projects. The page lists topics such as the Cowork task loop, plugins and skills, file and research workflows, steering multi-step work safely, using Claude in Chrome and Microsoft 365 apps, and sharing work with your team. Its curriculum has 15 items across five sections, including a final quiz.
I've included it because it's aimed at everyday knowledge work, not at programmers. Many people use AI only through a chat box, one question at a time. Learning to hand over a whole task with files, and then check the result, is a real step up in what you can do.
The section on steering multi-step work safely is the one to pay attention to. When a tool can act on your files and browser, you need to know how to give clear boundaries, review what it changed, and notice when it has gone off course. Those habits transfer to any agent product.
If you use a different assistant, the same habits apply, so compare this with your own vendor's material.
5. DeepTutor
DeepTutor is different from the others here, since it's a tool and not a course. It's an open source personalized tutoring system: you give it your material and it helps you learn it. Its recent releases mention a content workspace you can point at any folder, Mastery Path modes, durable Reading, immersive YouTube learning, and learner and guardian accounts.
The idea is to swap passive reading for active practice. Most people highlight a PDF and forget it within days. Being quizzed on the same material and made to explain it sticks far better, and a good tutor system pushes you to do that.
You can run it with Docker, and the README's Get Started section walks through setup and the model keys you'll need. It's a large, fast-moving project with frequent releases, so expect some breakage. Use it for the textbooks and courses on this list. Load the lecture notes and ask it to test you, and you'll find out what you really understood.
6. AI Engineering from Scratch
rohitg00/ai-engineering-from-scratch
This is the deep end. AI Engineering from Scratch is a free, open source curriculum with 542 lessons across 20 phases, in Python, TypeScript, Rust and Julia. Its stated goal is for you to implement model internals, retrieval pipelines and agent runtimes, test them, inspect failures, and keep the code and evaluation results.
What sets it apart is the emphasis on evidence. Projects come with a grader: you work in stages, run a test script, and end up with artifacts such as an HTML timeline and a JSON regression report. One example has you build an agent trace debugger that separates overlapping child work from parent execution time. That's real engineering, well beyond following a tutorial.
Don't try to finish it. At 542 lessons, it's a reference to dip into. The README offers learning paths for model foundations, LLM systems, and agents and delivery, plus a study guide on prerequisites and time. Pick a path, build its projects, and keep the results in a public repo. A working project with an evaluation report tells an employer far more than a certificate.
Which path fits you?
- Non-programmer using AI at work: Introduction to Claude Cowork, then the prompt tutorial.
- Complete beginner who wants to code: CS50P, then AI for Beginners.
- Working developer new to AI: the prompt tutorial, then AI Engineering from Scratch.
- Student with a heavy reading load: DeepTutor alongside your classes.
A 12-week plan with one hour a day
Here is one way to spread this out if you're starting from scratch:
- Weeks 1 to 4: the first half of CS50P, doing all the problem sets yourself.
- Weeks 5 to 8: the rest of CS50P, and write one small script that solves a real chore of yours.
- Weeks 9 and 10: the prompt tutorial, with a small API budget.
- Weeks 11 and 12: build a small tool that calls a model, write tests for it, and publish it on GitHub.
That plan is only a guide, and you may need longer. Finishing one thing matters more than starting all of them.
How to learn with AI without fooling yourself
An AI assistant can make you feel like you're learning when you aren't. Reading a clear explanation feels like understanding, right until you try to do the thing alone. A few rules help keep the learning real:
- Try the problem yourself first, and ask only when you're stuck.
- Ask for a hint, not the answer.
- Explain your solution back to the AI, and ask it to find the gaps.
- Close the chat and rebuild the thing from memory the next day.
- Check facts against the official docs, because models get details wrong with total confidence.
Set a weekly schedule, put it in your calendar, and tell someone what you're doing. Free courses are easy to abandon, and often the reason is that nothing makes you show up.
Show your work
Certificates are nice, but public projects carry more weight. Each time you finish a section, build one small thing and write a short note on what broke and how you fixed it. A year of small notes makes a better portfolio than one big unfinished idea.
When you're ready to practice on real scripts, my paid Python Automation Scripts Pack has 100 practical projects you can study, change and make your own, which suits the stage right after CS50P.
What to skip, and what to ignore
Free material is everywhere, and the risk is collecting it instead of doing it. A saved list of fifty courses teaches nothing. Pick one, finish it, then pick the next.
Be careful with content that promises fast results. Ignore courses that claim you'll be an AI engineer in a weekend, or lists of secret prompts. Prompts are easy to copy, so they're worth little by themselves. What lasts is knowing how models fail, how to test their output, and how to break a vague goal into checkable steps.
Don't stress about which model or tool is the newest either. Tools change every few months. Programming basics, clear writing and careful testing have stayed useful for a long time, and they make each new tool easier to learn. If you only have time for one thing this month, write code by hand and test it.
Pay for learning only when free material has stopped being enough. That moment usually comes when you need feedback on your own work from someone experienced, and that's the one thing a free course can't give you. A mentor, a code review or a study group is often worth more than another video series.
Finally, find other learners. The CS50 community, the GitHub discussions on these repos and local developer meetups all help. Explaining something to another beginner is one of the fastest ways to find out what you don't know yet, and it makes the long middle of any course easier to get through.
Pick one course today and start the first lesson before you close this page.
Comments