Harvard's free courses went viral again. But they're just the starting point. Here's the full map, organised by who you are and what you actually need.
Every few months, a post circulates: "Harvard is offering free AI courses. Here's the link." Thousands of reposts. Hundreds of comments saying "saving this." And then nothing happens. People save it, they don't do it, and three months later the same post goes viral again.
I'm not here to guilt anyone. The real problem is that most people don't know which course to start with, what level they're at, or what "learning AI" even means for their specific situation.
So here's the full map. Not just Harvard. All of it. Organised by who you are and what you actually need.
We are in a narrow window. The gap between people who understand how AI works, even at a high level, and people who are still waiting to "see how this plays out" is widening faster than any skills gap I've seen in my working life.
This is not about becoming a data scientist. Most of the courses below don't require you to write a single line of code. What they give you is a mental model: how these systems actually work, where they fail, what they can't do, and where the real leverage is. That mental model is becoming table stakes for anyone in a knowledge-work role.
The executives I work with who understand this, even conceptually, make better decisions about AI investments, ask better questions, and don't get sold vaporware. The ones who don't are either over-trusting the hype or refusing to engage at all. Both positions are expensive.
Every major technology shift of the last 30 years had an 18-to-36-month window where going first compounded. The internet, mobile, cloud, each time. That window is open right now with AI. Not at the beginning, not at the end. But it won't stay open.
This is the one that keeps going viral, and for good reason. It's a proper introduction to the concepts behind modern AI: search algorithms, knowledge representation, machine learning, neural networks, NLP, and computer vision. You write actual Python. You build things.
Designed for people who already know some programming. If you can read code but don't consider yourself a developer, you'll be fine. The certificate costs money, but you don't need the certificate.
cs50.harvard.edu/aiA shorter, non-technical course designed for people who need to understand generative AI for strategic decision-making without needing to implement it. Good for leaders, managers, consultants. Search "generative AI" at pll.harvard.edu and filter for free courses.
pll.harvard.edu/catalog/freeFor business professionals, managers, consultants, and anyone who needs to understand AI conceptually and apply it to their work.
The best entry point for anyone starting from zero. Created by the University of Helsinki and Reaktor, it explains what AI is, how machine learning works, what neural networks are, and what AI can and can't do, all without any programming. Clear, well-structured, doesn't talk down to you.
Two parts: "Elements of AI" (foundations) and "Building AI" (slightly more technical, optional). The first part is the one you need. About 15 hours total.
elementsofai.comAndrew Ng created this specifically for non-technical people. It covers what AI actually is (not the Hollywood version), how AI projects work inside companies, how to spot opportunities and risks, and how to think about AI strategy. About 6 hours. Highly rated.
This is the one I'd hand to a C-suite executive or a board member who wants to stop nodding along in AI conversations and start having a real opinion. Free to audit on Coursera.
deeplearning.ai/courses/ai-for-everyoneGoogle's practical onramp to using AI tools at work. Covers writing effective prompts, using AI to boost productivity, understanding risks and limitations, and integrating AI into everyday work tasks. Self-paced, no technical background required.
Good if your goal is "I want to use AI tools better right now" rather than "I want to understand how AI works." Both are valid goals.
skills.googleA more recent version of Ng's thinking, updated for the generative AI era. Covers large language models, image generation, AI in business workflows, limitations, and the near future. More current than "AI for Everyone" and more focused on the tools you're actually encountering today.
deeplearning.ai/courses/generative-ai-for-everyoneSometimes you don't need a course. You need to learn a specific thing. These are all free, short, and immediately applicable.
Co-created by OpenAI and DeepLearning.AI. Teaches you to write effective prompts, build basic pipelines using the API, and think systematically about how to instruct language models. About 1.5 hours. Free, no subscription needed.
deeplearning.ai/short-coursesThe best library of practical, tool-specific AI courses available. Courses on building with LangChain, RAG systems, fine-tuning models, building agents, evaluating LLM outputs, and more. Each course is 1-2 hours. Many are free while in early access.
This is where I go when I need to learn a specific thing quickly. The quality is consistently high because they bring in the people who actually built the tools.
deeplearning.ai/short-coursesOriginally built to train Google engineers. Covers machine learning fundamentals, neural networks, fairness, and responsible AI. More technically rigorous than the beginner courses, but still accessible to people who aren't daily programmers. Significantly refreshed in recent years.
developers.google.com/machine-learning/crash-courseFor people who can code, or who want to get to the point where they can build AI-powered applications. These require more time and more tolerance for frustration, but they're the real thing.
Hugging Face built the infrastructure that most AI applications run on top of, and their free courses are some of the best technical resources available. The LLM Course covers transformers, fine-tuning, RLHF, building with the Hugging Face ecosystem, and deploying models. If you want to understand how the underlying systems work and start building on top of them, this is your path.
huggingface.co/learnJeremy Howard's fast.ai course has trained more working AI practitioners than almost anything else outside of university programmes. The philosophy is top-down: you build real things first, then learn the theory as needed. 9 lessons of about 90 minutes each. Requires Python, but not a PhD.
This is the honest deep end. If you finish this course, you'll be able to build things that actually work.
course.fast.aiFor the genuinely curious who want to understand how AI agents learn through trial and error. This is how AlphaGo works, how AI systems learn to navigate environments. Not immediately applicable to most business problems, but foundational for understanding where AI agents are heading.
huggingface.co/learn/deep-rl-courseThe learning path is free on Microsoft Learn. Covers AI workloads, machine learning concepts, and Azure-specific AI services. Useful if you're working in a Microsoft environment, which most enterprises are. The AI-900 exam costs money and is retiring June 2026 — the learning materials stay.
learn.microsoft.comHere's the honest thing I keep wanting to say when these "Harvard free courses" posts go viral: the course is not the bottleneck. The discipline to actually finish something is the bottleneck. And most of these platforms know it — they're designed around the fact that most people will sign up and disappear.
I'm 50. I've watched every major technology transition of the last 30 years. The internet, mobile, cloud, social media. Each time, there was a window, maybe 18 to 36 months, where the people who understood the new thing got a meaningful advantage. Not because they were smarter. Because they decided to go first and accept the awkwardness of being a beginner.
We're inside that window right now with AI. Not at the beginning — that was 2023. But we're not at the end either. The people building fluency with these tools today are going to look prescient in three years. The people who kept waiting for it to "stabilise" are going to spend a lot of time catching up.
My honest recommendation: don't try to "learn AI." Pick one problem you have at work right now. Find the shortest course that helps you solve it. Finish that course. Apply it. Then pick the next problem. The goal is a working relationship with these tools, not a certification.
The courses above are free. The only thing they cost is the time you were already going to spend saving links you'd never open.
If you're going to start somewhere, start with Elements of AI if you're non-technical, or CS50 AI if you can code. Both are genuinely good. Both are free. Both will change how you read every article about AI that comes across your feed.
That shift in perspective is worth more than any individual course completion. Once you understand what these systems actually are, the hype becomes legible, the risks become concrete, and the opportunities become clearer. That's the real value of education here. Not the certificate. The new set of eyes.
3DH Consulting publishes deep guides and daily signals on AI for people running real businesses. No fluff, no hype, no reposts of things that went viral three months ago.
Subscribe free Work with Pedro