How to learn AI as a beginner: from prompting to AI agents

To learn AI as a beginner without being a developer, start by using an assistant such as ChatGPT, Claude or Gemini on the real tasks in your week, not with theory or code. The core skill is called prompting: giving context, a role, examples, constraints and a format, then refining and checking. Automating work with AI agents comes next and then, if you want, programming.

This guide is for adults who want to use AI seriously in 2026: a shop owner in Tiznit, an office worker at a small Agadir firm, a teacher, a student.

Learning AI: what are we actually talking about?

"Learning AI" covers two different jobs. The first is using generative AI tools to work better: drafting, summarising, analysing, automating. The second is building those tools: mathematics, statistics, machine learning, a long path that usually runs through university. This guide covers the first, in three levels that build on each other.

Level 1: using an AI assistant well (prompting)

An AI assistant runs on a language model trained on vast amounts of text: it produces, word by word, a plausible answer. It does not "know" things the way an encyclopaedia does, so it can write an answer that is fluent, confident and wrong.

What you get out depends mostly on what you put in. A good prompt reads like the brief you would give a bright intern who knows nothing about your business.

Six habits of a good prompt

1. Give context: who you are, who the result is for, what it will be used for.
2. Assign a role: "You are a sales adviser writing for Moroccan customers."
3. Show an example: paste a message that worked, to set the tone.
4. Set constraints: length, language, what must never be promised.
5. Ask for a format: a table, a list, three short versions, a draft email.
6. Iterate and check: ask for specific corrections, then check figures, names and dates yourself.

Example: before and after

Weak prompt: "Write an ad for my shop."

Strong prompt: "I run a silver jewellery shop in the medina of Tiznit. My customers are mostly local families and tourists. Write three versions of a Facebook post, 60 words maximum, announcing a new collection of bracelets. Warm tone, no exaggeration, no prices. One version in French, one in Arabic, one in English. End by inviting people to message us on WhatsApp."

The second request takes a minute longer to write and often saves several rounds of back and forth. Keep your best prompts in a document: that library will become your most useful tool. As for languages, assistants are getting better at written darija, but quality varies: test before relying on it with customers.

Level 2: automating tasks with AI agents

An assistant answers a question. An AI agent receives a goal and acts: it breaks the work into steps, uses tools (reading an email, opening a spreadsheet, filling in a document), checks the result and carries on. Think of a fast, tireless intern who needs precise instructions and a manager who reviews the work. Mainstream assistants increasingly plug into email and spreadsheets, and no-code platforms chain these steps without programming.

Four useful automations for a small business in Morocco

• Draft WhatsApp replies: from a document listing your opening hours, prices and delivery times, the agent drafts answers to frequent questions. You read, adjust and send. WhatsApp Business already offers saved quick replies; AI tailors them to each message.
• Summarise documents: a tender specification, a lease, an administrative circular. Ask for a five-point summary, then a list of dates and obligations.
• Extract invoice data: supplier, date, amount before tax, VAT and total, into a spreadsheet. Check the totals, especially on scanned invoices.
• Draft quotes: the agent works from your price list and the customer's request. You approve every line before it goes out.

The golden rule: a person approves

An agent that gets something wrong gets it wrong fast, and at scale. Start with tasks where a mistake is cheap, and never let an agent send, pay for or delete anything without your approval. Test it on twenty or so real cases before you trust it.

Level 3: building with Python and APIs

This optional level suits people who want to create their own tools or aim for a technical career. The most direct route is Python, a readable language widely used in AI: variables, lists, loops, functions, reading an Excel file. Then come APIs, counters through which your program queries a service, such as an AI model, and gets a response back. With those two building blocks, a script can process a whole folder of invoices in one go.

To understand the ideas without coding, the free online course Elements of AI, created by the University of Helsinki and MinnaLearn, needs no programming for its first part. For children, the way in is usually play: see our article on robotics and coding for kids.

A 30-day plan to learn AI as a beginner

Allow 30 to 45 minutes a day, on your own tasks, with non-confidential or anonymised documents.

At the end of the month, ask yourself which task now takes you less time, at the same quality or better. If the answer is "none", go back to days 4 to 14: the problem almost always lies in the instructions. If you are aiming for level 3, swap days 23 to 30 for an introduction to Python.

Limits and risks you need to know

Hallucinations: check everything

A "hallucination" is an invented answer presented as fact: a law that does not exist, a misquoted reference, a wrong figure. The more specific, local or recent the topic, the higher the risk. Every figure, date, name or quotation gets checked against an official source before you use it.

Personal data: Law 09-08 and the CNDP

In Morocco, Law 09-08 protects individuals with regard to the processing of personal data, meaning any information that can identify someone: name, phone number, national ID (CIN) number, photo, voice. The CNDP, the authority that oversees the law, stated in a press release of 18 March 2025 (in French) that AI processing which uses personal data falls under this law.

Pasting your client file or your pupils' report cards into a public assistant means handing that data to a third-party service, often hosted abroad. The law regulates both disclosure to third parties and transfers outside Morocco. Good habits:

• Anonymise: replace names with "Client A", remove phone and CIN numbers.
• Check whether the tool uses your conversations for training.
• For professional use, prefer a business plan with written commitments on data.
• If in doubt, check the CNDP website or ask a legal adviser.

These pointers are not legal advice. Even without personal data, unpublished figures, contracts and passwords have no place in a public tool. If security interests you, see our guide to getting started in cybersecurity in Morocco.

Students: academic honesty

Using AI to understand a lesson, quiz yourself or get feedback on a draft is a good idea. Handing in work written by AI as your own is cheating, and you learn nothing. Follow your institution's rules and declare your use when asked. To revise, ask AI for questions rather than answers: that is retrieval practice, covered in our article on effective study methods.

How to choose an AI course in Morocco

"AI training" ranges from a one-evening talk to programmes lasting several months. To sort the offers:

• You work on your own cases: your anonymised documents and tasks, not just generic examples.
• Practice comes first: one workstation each, exercises in every session, few slides.
• You leave with a result: a prompt library and at least one automation that runs.
• The trainer talks about limits: errors, verification, personal data.
• Groups are small and promises reasonable: be wary of "AI expert in two days".

A training centre certificate records your course; it is not a state diploma. And a paid course is not compulsory: with discipline, the plan above and free resources are enough to get going.

In Tiznit, the ApexTech programme at ApexTiz includes an AI track for professionals: prompting, autonomous agents and automating business tasks. Groups are capped at 10, with one workstation each, on site or remotely, and trainers explain in darija or French. The aim: leave with at least one automation running on your own data. For office jobs, the AI at work course applies these levels over four mornings: rules, AI in the browser, connectors, small agents and a final project.

Frequently asked questions

Can you learn AI without knowing how to code?

Yes. Using an assistant and automating with no-code tools need no programming. Code becomes useful when you want to build your own tools.

How long does it take to learn to use AI?

A few weeks of regular practice is enough to get comfortable with prompting. Reliable automations usually take a few months of trial and error. Becoming an AI engineer takes several years of study.

What is the difference between a chatbot and an AI agent?

A chatbot answers your messages, one exchange at a time. An agent receives a goal, uses tools such as email or a spreadsheet and chains several steps together. In both cases, a person checks the result.

Can I use ChatGPT with my clients' data?

Not without precautions. Law 09-08 regulates how personal data is processed and transferred: anonymise before pasting anything and, if in doubt, consult the CNDP or a lawyer.