Every AI roadmap you have found is a list of forty courses. You bookmarked it, finished two, and quit. The problem was never your discipline — it was the shape of the plan.

How to learn artificial intelligence: pick one layer, read 25 pages a day, and ship one small artifact every week. Most learners need only three layers — using AI tools well, data literacy, and evaluation. Add Python and math only if you are heading toward building models.

What do you actually need to learn about AI?

Roadmaps overload you because they are written for one reader: a future machine learning engineer. You are probably not that reader. A marketer, a fifth-grade teacher, and an aspiring practitioner need three different columns, and only one of them needs linear algebra.

Here is what the overload looks like versus what actually moves you:

Breadth-first learning ages badly here. Corporate AI investment hit $252.3 billion in 2024 according to the Stanford HAI AI Index, so the tool list changes every quarter. The layers underneath do not.

The seven AI skill layers (and who can skip each one)

Notice the shape: the first two and last one are universal. The middle four are role-specific. That is the whole reason a single roadmap fails three readers at once.

Three 90-day AI learning roadmaps

Each path is one vertical column. Twenty-five pages a day, one artifact a week, no parallel subjects. Pick the one that matches your job, not the one that sounds most impressive.

Path A — Non-technical professional who wants to use AI

1. Days 1–30: Prompting as a craft. One tool only. Build a personal prompt library of ten patterns for your actual tasks — summarizing, drafting, rewriting, comparing, extracting.
2. Days 31–60: Data literacy. Learn to read the tables you already receive at work. Practice rewriting one misleading chart per week in plain language.
3. Days 61–90: Verification and workflow. Learn where models fail, build a two-step check for every AI output you send to a human, and document one team workflow.

Skip Python. Skip calculus. In 90 days you have a portfolio of real work.

Path B — Teacher who must teach with and about AI

1. Days 1–30: Your own fluency. Same prompt-library work as Path A, but framed around lesson planning, differentiation, and feedback.
2. Days 31–60: AI literacy for students. Work through a competency framework rather than inventing one. UNESCO launched AI competency frameworks for both students and teachers at Digital Learning Week 2024, built around a human-centred approach, critical thinking, and responsible use.
3. Days 61–90: Policy and assessment. Write your classroom AI policy, redesign two assignments to be AI-resistant, and build one lesson that teaches how a model actually predicts.

You are not behind. In the AI Index education chapter, 81% of computer science teachers agreed AI belongs in a foundational CS experience, while fewer than half of high school CS teachers felt equipped to teach it. Ninety days of vertical work puts you in the prepared half. For the wider habit, see Professional Development for Teachers.

Path C — Aspiring practitioner heading toward machine learning

1. Days 1–30: Python plus tables. Syntax, functions, files, then pandas. The pandas getting-started guide is organized exactly the way work arrives: read a CSV, select subsets, derive columns, summarize, plot.
2. Days 31–60: Machine learning concepts on real data. Fit, predict, cross-validate, and build a pipeline with scikit-learn, whose estimators all expose the same fit and predict methods — one API, dozens of models.
3. Days 61–90: Build with an API, then evaluate it. Ship one small application, then test it properly. OpenAI’s evals guide describes the pattern in three steps: describe the task, run it against test inputs, and check the results against criteria.

Math runs alongside, not instead. Deep learning sits on algebra and calculus, so the 25Pages Mathematics column (Algebra → Calculus IV) is the vertical that makes Path C durable rather than memorized: Algebra for notation and functions, Calculus I–II for derivatives and gradients, Calculus III–IV for the multivariable thinking behind optimization. The HTML coding path feeds the same route from the other side — MDN’s HTML module requires no previous coding knowledge and teaches structure, forms, and semantics, which is exactly the interface layer your model needs before anyone else can use it.

What should I build each week?

The artifact is the point. Twelve weeks, twelve things that exist.

How do I avoid the tutorial trap?

The tutorial trap is watching, nodding, and mistaking recognition for skill. Two findings should reorganize your study time. In PNAS, Freeman and colleagues found active learning raised exam and concept-inventory scores by an average of 0.47 standard deviations, and students in traditional lecture courses were 1.5 times more likely to fail. In Science, Karpicke and Blunt found retrieval practice produced more learning than elaborative studying with concept mapping — even when the final test was building a concept map.

Translated into a protocol you can run today:

1. Read 25 pages. Stop at 25 even if you want more.
2. Close the tab. Write from memory what you just learned, in your own words.
3. Watch the idea executed once, then immediately imitate it yourself — that is Watch&Learn®.
4. Break the example on purpose. Change one input and predict the output before you run it.
5. Explain it out loud to someone who does not care about AI.

If you cannot do step 5, you watched a tutorial. You did not learn. More on the mechanics in How to Study Effectively and How to Learn Fast.

How do I prove I actually learned AI?

Nobody can inspect your course history. They can open your artifacts. Certificates come from proof, not attendance — so pair them.

• Portfolio first. One page, twelve artifacts, each with a two-line note on what you decided and what you would change. Include the failures; they read as judgment.
• A safety section. Name the risks in your own project using shared vocabulary from something like the NIST AI RMF, which frames trustworthiness across design, development, use, and evaluation.
• The certificate as an index. Use a 25Pages certificate on your résumé and LinkedIn as the label on the column you finished — then let the artifacts carry the argument.

FAQ

How long does it take to learn AI? Ninety days of 25 pages a day gets a non-technical professional to fluent use and a beginner practitioner to a working, evaluated project. Mastery of the practitioner column takes longer because math runs underneath it.

Can I learn AI without coding? Yes, for Paths A and B. Prompting, data literacy, and evaluation require no code. You only need Python once you want to train or deploy models yourself.

Do I need math to learn AI? Not to use it. Yes to build it well. Start at Algebra and stack upward to Calculus IV; skipping straight to deep learning math is why most learners stall in month two.

What is the best AI learning roadmap for beginners? The one with a single column and a weekly artifact. Any roadmap listing more than three subjects at once is a reading list, not a plan.

How should teachers start learning AI? Build your own fluency for 30 days first, then move to student-facing AI literacy using an established competency framework, then write your classroom policy. Teaching with AI before teaching about it never works in reverse.

Is a certificate worth anything in AI? Only alongside proof. A certificate tells a hiring manager which column you finished; the portfolio shows they can trust it.

Start today

Pick your path — A, B, or C — and read the first 25 pages before you close this tab. Begin at 25pages.com.

Read next

• Best Homeschool Learning Methods
• Professional Development for Teachers That Actually Changes Practice
• Lifelong Learning Programs
• How to Improve Emotional Intelligence
• How to Study Effectively
• How to Learn Fast

25Pages is a learning platform where you read 25 pages a day and master one idea better than anyone else. Watch&Learn® · Vertical Learning · 25pages.com · ECIS Group LLC