For most beginners, learning AI does not mean programming but using an AI chat such as ChatGPT, Claude or Gemini safely for their own work. As a rule of thumb: a basic understanding in one to two hours, then one tool, one real task a day and, after two to three weeks, a first project of your own. There are plenty of free courses, for example Elements of AI or the academies of Anthropic and OpenAI, but they do not replace practising on your own tasks.
I have worked with Claude every day since February 2025. My fourteen AI employees run on Claude Team, some also on open models or through the API. In this article I show you a path that works without prior knowledge and costs nothing. The course offers were checked against the providers’ websites for this article, with sources current to 31 October 2026. I collect more basics about AI at work on my AI Basics page.
What "learning AI" means for you
People mean two very different things by "learning AI":
- Using AI: You use ready-made tools to write texts, summarise documents or simplify routines. You need neither maths nor programming for this. This is the path for freelancers, offices and small teams, and it is what this article is about.
- Building AI: You build models or applications yourself. That takes programming, statistics and a lot of time. It is a career, not a starting point.
In between there is a third area that becomes interesting later: connecting AI with other programs, meaning automating routines. You get there once the basics are in place.
Learning AI step by step
Stage 1: understand what AI is
One to two hours are enough to start. You should know that an AI chat is based on a language model that predicts text word by word and does not look up whether something is true. Almost everything else follows from that: why the answers sound fluent, why they can still be wrong and why your material matters so much. How this works in detail is in How Do LLMs Work? A Plain-English Guide. What to keep in mind at the start is in AI for Beginners: What to Know First.
Stage 2: choose and set up one tool
Pick a single tool and stick with it for two weeks. For a beginner, the difference between the big providers is smaller than what you gain by not switching all the time. On day one, set the privacy switch so your chats are not used for training without you deciding. With Claude, Claude Guide: Getting Started walks you through the first steps.
Stage 3: one real task every day
This is the most important stage. Give the AI one task from your work every day that comes up often and carries little risk: a reply to a customer enquiry, a summary, an outline. Explain the task the way you would explain it to a new employee. Ready-made requests to try are in Prompt Templates: 12 Copy-and-Paste Briefs for Small Businesses. What makes a good request is in What Is a Prompt? Prompts and System Prompts.
Stage 4: know the limits
After a few days you will notice where the AI gets things wrong. Learn the signs: made-up numbers, sources that do not exist, long chats in which early details get lost. How to spot made-up answers is in AI Hallucinations: Why AI Makes Things Up. Why long chats get worse is in Context and Context Windows in AI Explained. This also covers what does not belong in the chat; my three questions are in AI and Privacy: What the AI Gets to See.
Stage 5: a project of your own
After two to three weeks, pick a task that regularly costs you time and build a fixed solution for it: a saved instruction, a project with your documents, a template you use again and again. This is where you learn the most, because you find out what the AI needs to work reliably.
Free courses to get started
A course helps if you want a structured approach to stage 1 or a record of completion matters to you. According to the providers, these offers are free (as of October 27, 2026):
Two notes:
- Vendor courses teach their own tool. That is handy once you have decided. For general understanding, Elements of AI is more neutral.
- Badges are not certifications. OpenAI itself states that its badges and certificates of completion are not certifications. And for the EU's AI literacy duty at work, the law does not require any particular certificate, more on that below.
If you work with Claude, Learn Claude: The Path in Five Stages shows the path for that tool specifically. Whether a paid course is worth it for you, and what has to run in daily work afterwards, is in After the AI Course: What Runs Every Month.
An example from everyday office work
A made-up but typical example: a tax assistant wants to learn AI because his boss announced it. One evening he goes through the introduction at Elements of AI and sets up an AI chat the next day. For two weeks he has the AI draft one client email every morning, without real names, and compares the draft with what he would have written himself. In the third week he builds a fixed instruction for follow-up questions about missing receipts. In the end he can show his boss what works and where the AI makes mistakes. He learned mostly from his own emails.
What to watch out for
Not too much at once
If you test five tools at the same time, you learn none of them properly. One tool, one task a day, that is enough for the first weeks.
Courses without practice fade
A course explains what is possible. What works in your daily work you only find out by trying it with your own documents. Plan at least one hour of practice on real tasks for every hour of course.
Checking results stays mandatory
Even after weeks of practice the AI can be convincingly wrong. Make it a habit from the start to check figures, names and quotes against the source.
AI literacy at work
As soon as people in your business use AI at work, Article 4 of the EU AI Act applies: you have to take measures to support their AI literacy. Germany's Federal Network Agency, which supervises this there, writes that no standardised training measures are prescribed and there is no certification requirement, but the measures should be documented. A self-organised learning path like this one can be part of it. This is not legal advice. What it means in practice for a small team is in AI Literacy Under the AI Act: Training Staff.
Frequently asked questions
Can I learn AI without prior knowledge?
Yes. Using AI chats requires neither maths nor programming. Elements of AI, for example, is explicitly aimed at people without such knowledge. More important than prior knowledge is that you practise regularly on real tasks.
How long does it take to learn AI?
One to two hours are enough for the basics. You will feel confident in daily work after two to three weeks with one task a day. That is a rule of thumb, not a fixed rule. Building AI yourself is a career and takes years.
Where can I learn AI for free?
For example at Elements of AI, the Anthropic Academy and the OpenAI Academy, and in German at the KI-Campus. According to their providers, all of them are free. On top of that, the free plans of the AI chats themselves let you practise right away.
What is the best way to learn ChatGPT?
The same as with any other tool: set it up, check the privacy setting, give it one real task every day and read the result critically. The OpenAI Academy offers free introductions for this, such as "AI Foundations".
Do I need a certificate?
Not for yourself. If your staff use AI at work, the EU AI Act requires measures for AI literacy but no particular certificate. Keep a note of what you have done. This is not legal advice.
Where to go from here
Set up one tool today and give it your first real task tomorrow morning. Which AI you can use for free and where the limits are is in Free AI: What the Free Plans Can Do. The key rules for getting started are summed up in AI for Beginners: What to Know First. The next step in daily work is writing with AI. Look up technical terms you come across in AI Terminology: A Plain-English Glossary. We discuss beginners' questions in my community AI Employee Framework.
Your first AI employee up and running within an hour
Join the community$5/month founding price, first 50 members
