Artificial Intelligence A-Z: Build an AI — Review

Our review

Artificial Intelligence A-Z is a focused, project-driven dive into reinforcement learning — the branch of AI behind game-playing agents and self-driving simulations — rather than the broad survey its title might suggest. Over 17 hours, de Ponteves and Eremenko take you from Q-learning fundamentals to building a deep Q-learning agent, culminating in AIs that steer a simulated self-driving car and play Doom.

The course is at its best when it makes an intimidating topic tangible. Reinforcement learning has a reputation for heavy maths, but the instructors lead with intuition — exploration versus exploitation, reward shaping, how an agent learns from consequences — before showing the code in PyTorch. Watching an agent go from random flailing to competent play is genuinely motivating, and it cements the concepts better than equations alone would.

This is the right course for someone specifically drawn to reinforcement learning and building agents, and for learners who already have some Python and want a hands-on, memorable project rather than another regression tutorial. The self-driving and Doom builds are the hook, and they deliver a real sense of "I made an AI that learned."

The honest limitation is scope and prerequisites. Despite the A-to-Z name, this is not a broad AI course — it concentrates on reinforcement learning and largely skips the supervised-learning breadth most people need first. It also assumes basic Python and some comfort with maths; a true beginner will struggle. The 4.4 rating, a touch below its stablemates, partly reflects learners arriving expecting general coverage and finding a specialised one.

For that broader foundation, the obvious companion from the same instructors is Machine Learning A-Z, which covers the supervised-learning toolkit this course assumes, and Deep Learning A-Z for neural networks in depth. If you want the underlying theory for free first, Andrew Ng's Machine Learning Specialization is the canonical grounding.

A note on managing expectations, since the rating dip is instructive: read the syllabus carefully and come for reinforcement learning specifically. If that's what you want — building agents that learn from reward — this is one of the most engaging on-ramps available, and the self-driving and Doom projects are genuinely fun. If you wanted a broad AI survey, you'll be disappointed not by the quality but by the scope, so match your goal to the course before buying.

Verdict: at $13.99 this is great value if reinforcement learning is what you're after — a vivid, hands-on introduction to a genuinely hard topic. Come in with Python and some ML exposure, treat it as a specialist course rather than a survey, and pair it with a broader ML course for balance. Our editorial score: 9/10.