How to get into AI with zero experience: what to actually do in your first 30 days

TL;DR: Do not start with courses, maths, or Python. Start by using AI on the work you already do, every day, for a month. Depth comes from repetition on real tasks, not from watching someone else explain transformers. Thirty days of daily use will put you ahead of most people who spent the same month collecting bookmarks.

The mistake almost everyone makes first

You decide to “get into AI”, so you look for a course. You find a 40-hour one covering neural networks, gradient descent and Python libraries. You do six hours, understand roughly a third, and quietly stop.

The problem is that you learned about how models are built, when what you needed was skill at using them. These are different disciplines with almost no overlap, and for the overwhelming majority of people — including most people who end up with “AI” in their job title — only the second one matters.

Think of it like the internet in 1999. Almost nobody who benefited from it learned TCP/IP. They learned to use it well, early.

Week 1: Use it daily on real work

One rule this week: every day, use AI on something you were going to do anyway. Not a practice exercise. Real work with real stakes.

Pick one paid assistant and stay with it for the month — Claude or ChatGPT, both around $20/month. Switching between tools while learning means you never get good at either.

Seven starting tasks, one per day:

1. Rewrite an email you are struggling with.
2. Summarise a long document you needed to read anyway, then skim the document and check what the summary missed.
3. Explain something you do not understand from your own field, then ask three follow-up questions.
4. Turn messy notes into a structured document.
5. Draft something you have been avoiding. Anything. The blank page is where this helps most.
6. Ask it to argue against a decision you have made.
7. Plan something — a trip, a project, a difficult conversation.

By day seven you will have formed an accurate mental model of what it is good at, which no article can give you. Keep a running note of what worked and what disappointed you.

Week 2: Learn to ask properly

Now that you have felt the difference between a good and bad response, learn why. Four techniques, one every two days:

• Give it a role and an audience. “You are a sceptical technical reviewer. My reader is a busy operations manager.” Changes everything.
• Show, do not describe. Paste two examples of what you want. Examples outperform adjectives by a wide margin.
• Constrain the output. Length, format, what to avoid. “Under 150 words, no bullet points, no preamble.”
• Ask for critique instead of production. “Do not rewrite this. Tell me the three weakest parts and why.” This is the single highest-value habit and almost nobody does it.

Then write yourself a system prompt — custom instructions describing who you are and how you want responses. Ten minutes that improves every conversation for months. The broader skill is covered in the prompt engineering piece, which also answers honestly whether it is worth learning as a career.

Week 3: Understand just enough theory

Now — and only now, because it will make sense against experience you actually have — learn the concepts that explain the behaviour you have been seeing. Roughly one short read each:

• Tokens — why long chats get expensive and slow.
• Context windows — why it forgets the start of a long conversation.
• Hallucination — why it invents things confidently, and what to always verify.
• Training vs inference — why it does not know last week, and whether your data trains it.
• RAG — how it works with your own documents.

That is genuinely enough theory to be competent and to stop being fooled. You do not need the maths. If you later find you need it, you will know exactly why, which is a far better reason to learn it than obligation.

Week 4: Build one thing that saves you time every week

Theory without application evaporates. Spend week four building a single working thing:

• A personal assistant configured for your job — role definition plus context files. Walkthrough in the assistant build guide. Best default choice.
• A searchable collection of your own documents — see the knowledge base guide. Best if you drown in files.
• One automation that runs weekly — a digest, a follow-up catcher, a report draft.
• Something small and slightly technical, if you are curious — a script that does one useful thing. You do not need to be able to write code; see what vibe coding actually gets you and the honest account in building an app as a non-developer.

Pick one. Finish it. A finished small thing teaches more than four abandoned ambitious ones.

The single exercise that teaches the most

If you do one thing from week one, do this — and do it on a topic you already know well.

Ask a question in your own area of expertise, where you already know the correct answer. Something with real substance, not a definition. Then read the response critically and mark it up: what is correct, what is subtly wrong, what is confidently stated and unsupported, what a novice would accept without question.

Almost everyone finds the same thing: the response is roughly 80% right, fluent throughout, and the wrong 20% is written in exactly the same confident register as the right 80%. There is no tonal signal separating them.

That single observation is the most useful thing you will learn in the entire month, because it generalises. When you later ask about something you cannot check, you now know the output looks identical whether it is right or wrong. Fluency is not evidence.

Then run the more useful version:

“Answer this question, then separately list which parts of your answer you are confident about and which you are uncertain about, and tell me specifically what I should verify independently and where.”

You will not always get a fully honest self-assessment — models are not reliable narrators of their own uncertainty — but it is substantially better than nothing, and it builds the habit that matters: treating output as a draft to check rather than an answer to accept.

People who skip this step are the ones who eventually forward something confidently wrong to a colleague. It is a rite of passage best avoided.

What to deliberately ignore for now

• Model release news. A new leading model every few weeks changes nothing about how you should work. You can safely ignore all of it for a year.
• Prompt packs and “1000 prompts” products. You will write better ones for your own work within a fortnight.
• Learning Python “for AI” unless you already want to program. It is a genuine skill with a real learning curve, and it is not on the path from here to competent user.
• Anyone promising an AI income in 30 days. Unfailingly, the person making money is the one selling the course.
• Agent frameworks and orchestration tools. Real, and several levels above where you are.

How to tell it worked

At day 30 you should be able to say: I use this daily without thinking about it, I can tell when the output is wrong, I know which tasks it is bad at, and I have one thing running that saves me time every week.

That is a genuinely valuable position — and it is roughly where the top decile of professional users are, because most people never get past occasional casual use.

What comes after is a real fork: deepening into your existing field, or moving toward AI-specific work. That is the next guide.