Skip to main content
Learning / Career
How to Learn AI
Whether you want to use AI tools more effectively, understand how they work, or build a career in AI — here's a structured roadmap for 2026 with the resources that actually work.
First: what do you actually want to learn?
The question "how do I learn AI?" needs to be sharpened before you can answer it. There are three very different goals people have, and the right path depends entirely on which one is yours:
AI literacy
Use AI tools effectively in your existing work
- Read: The AI Briefing newsletter for applied AI context
- Practise: Use ChatGPT or Claude daily for work tasks
- Learn: Google's "AI Essentials" free course (6 hours)
- Advance: Elements of AI (University of Helsinki) — free, no code
- Apply: Identify 3 recurring work tasks to AI-assist this month
AI application
Build AI-powered tools and workflows (some coding)
- Foundation: Python basics (CS50P on edX — free)
- APIs: OpenAI or Anthropic API quickstart tutorial
- No-code automation: Make.com or Zapier AI courses
- Prompting: Anthropic's prompt engineering guide
- Build: Ship one AI-powered tool using API + your use case
AI/ML engineering
Build and fine-tune AI models professionally
- Maths: Linear algebra and calculus (3Blue1Brown YouTube)
- ML theory: fast.ai "Practical Deep Learning" — free
- LLMs: Andrej Karpathy "Neural Nets: Zero to Hero" — free
- Framework: PyTorch fundamentals (official tutorial)
- Fine-tuning: Hugging Face course — free, hands-on
Best resources by category
Free courses (no coding required)
- Elements of AI — University of HelsinkiThe most accessible introduction to AI concepts. Covers what AI is, machine learning, neural networks, and ethics. Genuinely no-code. 40,000+ completions. Free certification.
- Google AI Essentials6-hour practical course covering AI tools at work, responsible AI, and prompt writing. Part of Google's Career Certificates. Free on Coursera (audit mode).
- Microsoft AI for BeginnersGitHub-based curriculum with 12 lessons, visual notebooks, and no heavy maths prerequisites. Good for the technically curious non-specialist.
Courses for technical learners
- fast.ai — Practical Deep LearningJeremy Howard's course is widely considered the best entry point to deep learning for practitioners — starts with code, explains concepts as they become relevant. Free. Assumes some Python.
- Andrej Karpathy — Neural Nets Zero to HeroYouTube series building a GPT-like model from scratch. Dense, technical, brilliant. The best way to understand how language models actually work if you're comfortable with Python and maths.
- Hugging Face NLP CourseHands-on transformer training, fine-tuning, and deployment using the Hugging Face ecosystem. Free. Practical. The fastest path to working with real models.
Paid courses worth the money
- DeepLearning.AI Short Courses (Andrew Ng)Series of 1-2 hour focused courses on specific topics: RAG, agents, fine-tuning, prompt engineering. $49/month for all. Best for filling specific knowledge gaps rather than structured learning.
- Udemy — Complete Machine Learning BootcampKirill Eremenko's course is outdated in parts but comprehensive for ML fundamentals. Often on sale for £12. Good for those who prefer video instruction over documentation.
Books for depth
- Hands-On Machine Learning (Aurélien Géron)The standard reference for ML engineering. Third edition covers deep learning and transformers. Dense but comprehensive. Best read alongside practical work.
- The Coming Wave (Mustafa Suleyman)Not technical — strategic and philosophical. Best book for understanding why AI matters at a societal level. Suleyman co-founded DeepMind and then Inflection AI.
- AI Snake Oil (Narayanan & Kapoor)Essential counterweight: which AI claims are real and which are marketing. Helps develop critical evaluation of AI capabilities claims.
The learning approach that works
The most common mistake is passive consumption — watching courses without building anything. Every AI concept you learn should be immediately applied to something real. If you learn about APIs, connect one to a tool you actually use. If you learn about prompt engineering, apply it to a problem you face at work. The people who learn AI fastest in 2026 are those who have a real problem they want to solve and use courses to find the pieces they need — not those who try to complete a curriculum before applying anything. Start with one concrete goal. Build something imperfect. Then learn the next piece.
FAQ
Do I need to know maths to learn AI?
Depends what level you want to reach. To use AI tools effectively at work: no maths required. To build applications using APIs and AI tools: basic Python is more useful than maths. To understand how models work conceptually: some linear algebra and statistics helps, but fast.ai has shown that starting with code and learning the maths when you need it is effective. To do original ML research or train models from scratch: yes — linear algebra, calculus, probability, and statistics are required. Most people learning AI for professional purposes fall in the first two categories, where maths is not a prerequisite.
How long does it take to learn AI?
It depends on your starting point and goal. AI literacy (using tools well): 2-4 weeks of daily practice. AI application (building with APIs, automation): 3-6 months of consistent learning and building. ML engineering (training models professionally): 1-2 years of structured learning and practice. The timeline accelerates dramatically if you have a real project — applying concepts immediately cements them far faster than course-only learning. The half-life of specific AI tools knowledge is short (things change fast), but understanding of fundamentals — what models can and can't do, how prompting works, how training shapes behaviour — remains durable.
Get AI insights every week
The AI Briefing covers what actually matters in AI — no hype, no jargon, just what you need to stay ahead.
Subscribe free