The short answer
To learn AI from scratch, work through six stages in order. First, understand what AI tools are and what they are not. Second, learn to brief them clearly. Third, learn to check what they give you. Fourth, apply them to one real task from your own work. Fifth, hand them bigger, multi-step jobs with a clear definition of done. Sixth, build a habit of keeping up that does not involve chasing every launch.
None of those stages needs coding, maths or a degree. That route is for people who want to use AI, which is most people asking the question. If you want to build AI systems yourself, the road is different and much longer: Python, statistics, linear algebra, machine learning and then deep learning. Both roads are covered below. Decide which one you are on before you start, because that single decision changes everything that follows.
First, decide: use AI or build AI?
Almost every "how to learn AI" roadmap you will find starts with Python and linear algebra. That is good advice for a future machine learning engineer and poor advice for a marketing manager, a teacher, a job seeker or a small business owner. They are answering a different question from the one most people are asking.
Ask yourself one thing: what do you want to be able to do in three months?
- If the answer sounds like "write reports faster", "handle my inbox", "prepare for interviews", "stop dreading admin" or "understand what my team is talking about", you want to use AI. Follow the first roadmap.
- If the answer sounds like "train a model", "work as a machine learning engineer" or "build AI products from the ground up", you want to build AI. Skip to the second roadmap.
- If you are not sure, start with using. Nothing you learn there is wasted if you switch later, and you will find out what these systems actually do before committing to years of study.
In plain English
- Using AI:
- Applying existing tools such as chat assistants to your own work: briefing them, checking them and fitting them into a routine. No code involved.
- Building AI:
- Writing software that trains, adapts or connects models. Needs programming, statistics and maths, and takes years rather than weeks.
- Large language model:
- The engine inside most AI assistants. It predicts likely text, which is why it can sound confident while being wrong.
If you are worried you lack some prerequisite, we answer that separately in what you need to know before learning AI. The short version: for the using route, you already have it.
The roadmap for using AI: six stages
Each stage below has four parts: what you learn, one exercise to do with a real task, the free LearnWithZavi course that covers it, and a rough sense of how long it takes. Do the exercise. Reading about a stage without doing the exercise is the most common reason people feel they "know about AI" but never use it.
Stage 1: Understand what these tools are and are not
What you learn. What generative AI does (it predicts plausible text, images or audio from patterns), why it can be fluent and wrong at the same time, what a model is, and why different tools feel different. You are building a mental model, not memorising definitions.
Exercise. Ask an AI assistant five questions about a subject you know well: your job, your town, your hobby. Mark each answer as right, partly right or wrong. You will see, in your own field, exactly how convincing a wrong answer can sound. That lesson sticks far better than any warning.
Course. Introduction to Generative AI, twelve short lessons. If you have never opened an assistant at all, Getting Started with ChatGPT or Getting Started with Gemini will get you through your first session.
How long. A sitting or two for the course. The exercise takes one coffee break.
Stage 2: Learn to brief them
What you learn. How to write a prompt that gets a useful answer the first time. A good brief says who the output is for, what it is for, what the assistant needs to know, what format you want and what to avoid. It is the same skill as briefing a capable new colleague who knows nothing about your situation.
Exercise. Take one real task from this week. Ask for it in a single vague line, then ask again with a proper brief. Put the two answers side by side.
โ Weak prompt
Prompt
Write an email about the project delay.
Output
A generic, apologetic email to an unnamed reader, with a vague new date and a paragraph of filler about valuing their patience.
The assistant had to guess the reader, the cause, the new date and the tone. It guessed all four, and you now have to rewrite most of it.
โ Good prompt
Prompt
Write a short email to a client who is expecting our website redesign on Friday. It will now be ready the following Wednesday because their product photos arrived late. Keep it calm and factual, do not over-apologise, confirm the new date, and ask them to approve the homepage draft by Monday so we stay on track. Under 120 words.
Output
A clear, polite email that names the real reason, gives the new date and asks for one specific action by a specific day.
Reader, reason, date, tone, action and length are all stated, so the first draft is nearly the final draft. The improvement came from your brief, not from a magic phrase.
Course. Prompt Engineering, twelve lessons from your first prompt to techniques like giving examples. Later, when long chats start drifting or forgetting your instructions, Context Engineering for Reliable AI is the next step.
How long. The course takes an evening or two. The skill takes a couple of weeks of briefing real tasks before it becomes automatic.
Stage 3: Learn to check them
What you learn. How to verify factual claims, how to spot invented sources and quotes, what you must never paste into a public AI tool, and how to recognise scams and fakes that use AI. This stage is what separates people who use AI well from people who use it a lot.
Exercise. Take one AI answer that contains facts, figures or references. Check three claims against a source you trust. Then write a short list of the information from your work that must never go into a public AI tool: client names, personal details, anything under contract.
Course. AI Safety, Privacy and Verification, six lessons. For the specific question of workplace data, read can I use AI with confidential work information?.
How long. A sitting or two. Checking itself never finishes: it becomes part of every task.
Do not skip Stage 3 because it feels less exciting than the others. An assistant that saves you an hour and slips one wrong figure into a report has not saved you an hour. Checking is a skill you learn, not a mood you are in.
Stage 4: Apply AI to your real work
What you learn. How to fit AI into the tasks that actually fill your week: emails, meeting notes, research, reports, summaries and planning. The goal is one workflow you trust, not twenty you tried once.
Exercise. Pick one task you do every week. Run it with AI on a low-risk example, keep your own version beside it, and note what was useful, what was wrong and what still needed you. Next week, run it again with an improved brief. Two runs of the same task teach you more than ten different experiments.
Course. AI Productivity at Work, six lessons that end with a weekly routine you can keep. There are also role-specific courses: see all courses for managers, small business owners, job seekers, educators and more.
How long. A few weeks of using it on real tasks. This is where most of the value lives, so do not rush it.
Stage 5: Delegate multi-step work
What you learn. The shift from asking AI questions to handing it whole pieces of work. That means writing a brief with a clear definition of done, giving it only the material it needs, knowing what an AI agent is (a tool that can take actions, not just produce text), and automating repetitive tasks without losing track of them.
Exercise. Choose one task where you want a complete first draft rather than a single answer, such as a briefing document built from three reports. Write the brief with a definition of done and a point where the assistant should stop and ask you. Review the result as if a junior colleague had produced it.
Course. Getting AI to Do the Work, then Introduction to AI Agents and No-Code AI Automation, six lessons each. If you want the background, our explainer on what an AI agent is covers the basics.
How long. Longer than the earlier stages, because you are designing a process as well as writing a prompt. Treat it as a month of occasional practice rather than a single sitting.
Stage 6: Keep up without chasing every launch
What you learn. How to stay current without burning out. New tools and features arrive constantly, and most of them do not change what you need to know. The durable skills are the ones from Stages 2 to 5: clear briefs, careful checking, good judgement about what to delegate. This stage also covers the career side: which human skills pair well with AI and how to keep them sharp.
Exercise. Set a monthly fifteen-minute review. Ask three questions: which AI task saved me real time this month, which one went wrong, and is there one new thing worth trying on a real task? If the answer to the last question is no, that is fine.
Course. Future-Proof Your Career with AI, six lessons on what AI is and is not replacing, and how to become the person who checks.
How long. Ongoing, but light. An hour a month is plenty once the earlier stages are in place.
The roadmap at a glance
| Stage | What you can do after it | Free course |
|---|---|---|
| 1. Understand the tools | Explain what AI is good at, where it fails, and why fluent is not the same as correct | Introduction to Generative AI |
| 2. Brief them | Get a useful first draft by stating audience, goal, context and format | Prompt Engineering |
| 3. Check them | Verify claims, spot invented sources and keep private data out | AI Safety, Privacy and Verification |
| 4. Apply to real work | Run one repeatable work task with AI and a human review step | AI Productivity at Work |
| 5. Delegate bigger work | Hand over multi-step tasks with a definition of done, and automate safely | Getting AI to Do the Work |
| 6. Keep up sensibly | Stay current without chasing launches, and build skills that pair with AI | Future-Proof Your Career with AI |
The roadmap for building AI
If you want to build AI systems, be honest with yourself about the size of the road. It is a technical career path measured in years, and the order matters.
- Python. The working language of the field. You need to be comfortable writing, reading and debugging real programs, not just following along with a tutorial.
- Statistics and probability. Distributions, sampling, uncertainty and how to tell a real effect from noise. Most of machine learning is applied statistics.
- Linear algebra and calculus. Vectors, matrices and derivatives. These are how models represent data and how they learn from it.
- Machine learning fundamentals. Training and testing, overfitting, evaluation, and classic methods such as regression and decision trees, practised on real datasets.
- Deep learning and large language models. Neural networks, how modern language models are trained and adapted, and how to evaluate them properly.
LearnWithZavi does not teach this route. Our courses are for people who want to use AI well, and we would rather say so plainly than pretend otherwise. If building is your goal, choose a developer-focused course or degree that is designed for it and follows roughly this order.
Even future builders benefit from the using roadmap first. A few weeks of Stages 1 to 4 will show you what these systems do well and badly in practice, which makes the theory far easier to care about.
How long does it take to learn AI?
For the using route: a few sittings to get useful, a few weeks of real practice to get comfortable, and a few months of everyday use to become the person others ask for help. For the building route: years. We go into the detail, goal by goal, in how long it takes to learn AI.
And yes, you can do all of the using route by yourself. If you want reassurance on that, read can you learn AI on your own?.
Common mistakes when learning AI
Tutorial hopping. Watching one more video or starting one more course feels like progress, but it is not practice. Finish one course, then use it on a real task before starting the next.
Starting with theory you do not need. If your goal is to use AI at work, calculus is not step one. It is a different subject. Starting there is the fastest way to decide AI "is not for you" when it very much is.
Chasing every new tool. A new assistant or feature every week is exhausting and mostly irrelevant. Pick one or two tools, get good with them, and let the rest come to you if they genuinely help.
Never checking the output. The most expensive mistake. Fluent text looks finished. Treat every answer as a draft from a fast, well-read colleague who has never met your client and sometimes makes things up.
Learning in the abstract. Practising on made-up examples teaches you the tool. Practising on your own real work teaches you where the tool belongs. Only the second one changes your week.
Where to find free resources
This post is the roadmap. If you want a wider look at free ways to learn, including what to ignore at the start, see how to learn AI for free.
Start today: the free 30-day path
If you want this roadmap turned into a daily plan, follow 30 Days With Zavi. Each day has one short lesson from our existing courses, one practical task and one small piece of evidence that you did it. There are 30 short lessons, no signup and no paywall.
It follows the same order as this roadmap. The first week covers what AI is, your first real request, how to brief well, and what not to paste. The middle of the month moves into email, meetings, research, reports and giving AI only the context it needs, then into delegation and simple automation. The final stretch covers customer messages, marketing, job search and portfolio work, and ends with a durable skills plan and a written plan for the next month. Day one asks you to write down two tasks AI could help with and one you would never hand over without review. It takes minutes.
If you would rather pick a route based on your goal, such as saving time at work, finding a job or running a small business, go to Start Here and it will suggest where to begin.
LearnWithZavi courses are free and run in your browser. Completion certificates are a record that you finished the lessons, not an accredited qualification. The real proof that you have learned AI is the work you can show at the end of the month.