The short answer
About twenty minutes to become genuinely useful, which is roughly one course on LearnAI. A few hours spread over a couple of weeks to get comfortable. A few months of using it on real work to become the person your team asks for help. Years, and a different career path, if you want to build the models yourself.
Those four answers are all correct, which is exactly why the question feels impossible to answer online. Someone promising to make you an AI expert in seven days and someone insisting it takes three years are usually answering different questions and not telling you which.
Why does nobody give a straight answer?
Because "learn AI" hides four different goals inside one phrase, and they are separated by orders of magnitude, not by a few weekends.
| Your goal | Realistic time | What it looks like |
|---|---|---|
| Use AI competently | 20 minutes | You can get a usable draft and you know to check it |
| Use it well at work | A few hours over a fortnight | You know the techniques and where it fails |
| Lead others in using it | A few months of real use | You know when not to use it, and you can explain why |
| Build the systems yourself | Years | Maths, code, data, a career change |
Most of the internet argues about the first row and the last row as though they are the same subject. They are not. Being fluent with AI tools has roughly the same relationship to machine learning engineering that being a confident driver has to designing gearboxes. Both are legitimate. Only one of them is what you asked about, and for most people it is the first three rows.
In plain English
- Machine learning engineer:
- Someone who builds and trains the models themselves. Requires programming, statistics and usually a few years of practice.
- AI literacy:
- Being able to use AI tools well, judge their output and know their limits. This is what most jobs actually ask for.
- Prompting:
- How you phrase a request to an AI tool. The single highest-return skill for a non-technical user.
What does twenty minutes actually buy you?
More than you would expect, and this is the part the "AI takes years" crowd gets wrong. In one sitting you can learn to open a tool, describe a task with enough context that the output is worth keeping, and understand that you must check what comes back. That is the whole entry fee.
Concretely, after twenty minutes you should be able to turn a vague request into a specific one. Not "write a marketing email" but "write a 120 word email to existing customers announcing a price change from April, apologetic but not grovelling, no exclamation marks." You should be able to look at the result, notice the two sentences that sound nothing like you, and fix them. And you should understand, in your bones rather than as a warning you read once, that the tool will state something false with total confidence and it is your job to catch it.
Those three things are most of the practical value of AI for most people. The rest is refinement.
Our courses run roughly 15 to 20 minutes each, with lessons averaging under 3 minutes. That is deliberate. The goal is that you finish one on a lunch break and use it that afternoon, rather than committing to a term of study. Getting Started with ChatGPT is the usual first stop.
What do a few hours buy you?
Roughly three to five courses, spread over a fortnight, gets you from "this works" to "I know what I am doing."
You pick up technique: giving the model a role, supplying examples, asking for a critique instead of a rewrite, breaking a big request into steps. You start to develop a nose for when the answer is likely to be wrong, which usually means anything involving recent events, specific numbers, legal or medical specifics, or claims about small organisations that barely exist online. And most importantly, you stop treating AI as a party trick and start pointing it at the boring things you do every week: the status update, the meeting notes, the same three emails.
That last shift is the one that actually changes your working life. Prompt Engineering covers the technique, and AI Productivity at Work covers the pointing-it-at-real-work part.
What do a few months buy you?
Judgement. Specifically, judgement about when not to use it.
This is the difference between a competent user and an enthusiastic one, and it is the only thing on this list that genuinely cannot be rushed, because it comes from having been burnt. You learn that AI is excellent for a first draft of something you know how to evaluate, and dangerous for a first draft of something you do not. You learn that it is faster to write the two-line reply yourself than to prompt for it. You learn which colleague will be quietly annoyed to receive machine-written text. You learn that for anything with real consequences, the tool goes at the start of the process and never at the end.
Nobody teaches you that in an afternoon. You get it by using AI on real work, noticing what went wrong, and adjusting. A few months of that and people start forwarding you things and asking how you would approach them.
Why do 30 day challenges usually fail?
They are not a scam, and the people who run them mean well. The format has a structural problem though: it front-loads motivation and back-loads application.
Day one is exciting. You are learning about transformers and tokens and the history of neural networks, none of which will help you next Tuesday. By day nine the novelty has gone, the material has not yet touched anything you actually do, and life intervenes. The application is scheduled for week four, and week four never arrives.
The version that works is smaller and less satisfying to announce: pick one real task you do every week and learn only enough to do that task with AI. Your weekly report, your client emails, your data cleanup. Learn that, use it four times, then pick the next task. Slower to describe, far more likely to survive a busy month.
If you want a challenge format, make the unit a task rather than a day. "Five tasks I have moved onto AI" is a much better target than "30 days of AI."
Why do some people take much longer?
Three reasons, and none of them is intelligence.
Tool hopping. Every week brings a new tool, and switching feels like progress because it involves clicking things. It is not. The core skill (describing what you want clearly and checking what comes back) transfers between every tool. Pick one, get good, then look around.
Watching instead of doing. You can spend forty hours on AI videos and be worse at using AI than someone who has spent forty minutes typing into one. Watching creates the feeling of learning without the friction that produces it. If you have not opened the tool during a lesson, you have not learned it.
Waiting to feel ready. There is no threshold. There is no point at which you have read enough to begin. The first prompt you write will be mediocre and that is entirely fine, because the second one will be better and it costs nothing.
A realistic plan if you have almost no time
Think in minutes per week rather than days on a calendar.
Week 1 20 min One foundation course. Then use it once, that day.
Week 2 20 min Prompting technique. Redo last week's task better.
Week 3 20 min A course matching your job. Apply it to one real task.
Week 4 0 min No new learning. Just use it on three real things.
Week 5+ 20 min One course a week, always followed by one real use.
That is about two hours of learning across five weeks and it will put you ahead of most people in most offices. Week four matters as much as the others: a week of pure application, with nothing new added, is where the material stops being information and becomes a habit.
There are 28 courses on LearnAI, all free, with no signup and nothing to cancel. If you are not sure where to begin, the start page picks a path based on your job, or you can browse all courses and choose the one that annoys you least.
Set a timer for twenty minutes. You will not be an expert at the end of it. You will be useful, which is the part that pays.