AI Learning Tools and Courses for Builders
The best AI learning resources split into three groups: programming foundations, AI-specific courses, and prompt craft. Start with Harvard CS50 if you need the basics. Move to fast.ai or Neural Networks: Zero to Hero once you can already write code. Use Learn Prompting if you only want to work with models, not build them. All 18 tools here are free or freemium, so the real cost is your time. Pick by what you want to have built in a month, not by prestige.
18 tools · Catalog updated
Education tools
Where do I start if I can already code?
Skip the fundamentals and go straight at the models. fast.ai teaches top-down: you train a working network in the first lesson and learn the theory underneath it later. That suits people who lose patience with math-first courses. Neural Networks: Zero to Hero goes the other way. Andrej Karpathy builds GPT from scratch in plain Python, line by line, so you finish knowing what every part does. It is slower and harder, and it is the one to pick if you want to modify models rather than call them. DeepLearning.AI sits between the two, with short courses on fine-tuning, agents, and RAG pipelines that assume you already know Python.
What if I need the programming fundamentals first?
Then do not start with AI at all. Harvard CS50 is the standard answer: C, Python, SQL, and web basics, taught as computer science rather than as a framework tour. It is free and it is hard, and it leaves you able to read code you did not write. The Odin Project is the alternative if you want to be shipping web apps sooner. It is project-based full-stack work in JavaScript, React, and Node, with no video-lecture format to sit through. Either path takes months, not weekends. Do one of them properly before you spend money on anything, because most AI courses assume Python fluency on day one.
Frequently asked questions
- Do I need to pay for any of these?
- No. Fourteen of the 18 tools in this category are free outright, and the other four are freemium, meaning the core material is free and payment unlocks extras like certificates or graded feedback. You can go from zero to fine-tuning a model without spending anything. Budget for compute instead, since training runs cost more than the courses do.
- Should I learn prompting or machine learning?
- Different jobs. Prompting is for building on top of existing models, and Learn Prompting covers the techniques for chat models, image models, and agents. Machine learning is for training and modifying models yourself, which is what fast.ai and Neural Networks: Zero to Hero teach. Most people shipping products need the first. Pick the second only if the model itself is your product.
- How do I stop abandoning courses halfway?
- Choose a format that matches how you actually work. Project-based curricula like The Odin Project give you something running at each step, which is easier to return to after a break than a lecture series. Video courses such as Harvard CS50 work better if you keep a fixed schedule. Either way, pick one and finish it before adding a second.