By Mufy Pachorawala, founder of getAIwork · Last updated: September 3, 2026
Key facts
- 68% of the 1,368 live listings on the getAIwork board are technical as of August 29, 2026, from 39,308 posts screened, but 27% are words-centred, so the entry point depends on which half you are aiming at
- Only 11% of live listings are explicitly beginner-friendly while 59% ask for advanced skills, which tells you how long the runway is
- The official study paths for every major vendor AI exam are free. You pay only for the exam, if you choose to sit one
- Paid AI training platforms screen with a qualification task, not a credential, so they double as a way to learn the domain while being paid, as listed by each platform
- You do not need mathematics to start. You need it to go deep, which is a different sentence and arrives much later than the curriculum sellers suggest
What should you learn first when learning AI?
Use the tools on work you actually have to do. Not toy prompts, not tutorials, real tasks with a real deadline where being wrong costs you something. Three months of that produces a kind of knowledge no course sells: an instinct for where models are reliable, where they quietly are not, and what a good prompt has that a bad one lacks.
This sounds too easy to be a curriculum, which is why most people skip it and start with a maths course they will abandon. The order matters because motivation is the scarce resource here. Tool use gives you results in week one and a reason to keep going in week nine. Linear algebra gives you neither, right up until the moment it becomes essential.
Concretely: run your writing, research, code review, data cleaning and planning through a model for a quarter. Keep a running note of everything it got wrong. That note is your syllabus, and it is better than any you can buy because it is about your work.
How long does it take to learn AI?
Depends entirely on what you mean by learn, so here is a table that says what each destination actually costs. These are ranges from what the work requires, not promises, and nobody can tell you where in a range you will land.
| Destination | Rough time | What you need | What it opens |
|---|---|---|---|
| Competent tool user | 1 to 3 months, part time | Consistent use on real tasks | Immediate value in your current role |
| Can build with APIs | 3 to 6 months | Python basics, one API, one small app | Automation work, prompt and eval roles |
| Can do AI training and evaluation work | Weeks, plus a qualification task | Careful reading, clear written justification | Paid platform programmes, as listed |
| Junior ML engineer | 12 to 24 months | Python, statistics, one framework, deployed projects | The technical 68% of listings |
| Research-adjacent | Years | Mathematics, papers, usually a postgraduate route | A small number of roles |
Most people asking how to learn AI want the first or second row and have been sold the fourth. That mismatch is where the wasted year comes from.
Do you need maths to learn AI?
Not to start. Yes to go deep. The useful version of this answer is about sequencing rather than yes or no.
You can use models, build with APIs, do evaluation work and ship useful automation with no mathematics beyond what you already have. You cannot understand why a training run diverged, design a loss function or read a paper properly without linear algebra, calculus and statistics. Both statements are true and people fight about them because they are answering different questions.
The efficient route is to learn the maths when a specific problem demands it. Hit something you cannot debug, find out it is a gradient issue, then go and learn about gradients with a concrete reason to care. Learning it in the abstract first has a famously high dropout rate, and the people who succeed at that route are usually the ones who already enjoyed maths, which is not most people.
Take the 2-minute match quiz and filter 1,368 screened AI listings to your current skills and level.
Take the 2-minute match quiz →
Can you learn AI for free?
Yes, and the free material at the top end is genuinely better than most paid courses. Vendor study paths for the major AI certifications are free. Several university courses are published in full. Framework documentation is written by the people who built the thing. Nothing you buy improves on that content.
What money buys is structure, accountability and a credential. Those are real products and some people need them. But be clear about which one you are buying, because paying for content that is free elsewhere is the most common way to overspend here. We break this down in free vs paid AI courses.
The free route’s actual weakness is not quality, it is sequencing. There is too much of it and no one tells you what to do next. Fixing that costs nothing: pick one path, finish it before starting another, and treat the abundance as a reason to commit rather than to browse.
What should you build to prove you have learned AI?
One thing, finished, deployed, that solves a problem someone actually has. Preferably a small problem. The single most persuasive artefact in this field is evidence that you completed something without anyone making you, because that is rarer than technical skill and much harder to fake.
Good candidates: something that automates a tedious part of your current job, a tool for a hobby community you belong to, an evaluation harness that compares two models on a task you care about. Bad candidates: another chatbot wrapper, a Titanic notebook, anything a cohort of forty people submitted the same week.
Write two paragraphs about what broke and how you fixed it. That write-up is often more useful in an interview than the project, because it is where judgement is visible. Anyone can follow a tutorial. Describing a failure accurately is the thing that separates people.
Can you get paid while learning AI?
Sometimes, and it is underused. AI training, annotation and evaluation platforms account for 634 of the 1,368 live listings on our board as of August 29, 2026. They screen with qualification tasks rather than credentials, which means the entry test is whether you can do careful work, not whether you bought a course.
The learning is real, not incidental. Spending weeks judging model outputs against guidelines teaches you how models fail in a way no course does, and that knowledge transfers directly into prompt work, evaluation roles and product roles. Pay is whatever each platform lists, and it varies a great deal.
Set expectations honestly: this is task work with unpaid qualification time up front and unpredictable volume, not a salary. Read AI training jobs with no experience and how the platforms differ before committing evenings to assessments.
1,368 live AI listings, screened and human-approved, filled ones deleted daily. Free and public, no signup.
Frequently asked questions
How do I start learning AI as a complete beginner?
Use AI tools heavily on real work for a few months and keep a note of everything they get wrong. That note becomes your syllabus. Starting with mathematics or a framework course before you have any practical feel for model behaviour is the most common way people quit.
How long does it take to learn AI?
Becoming a competent tool user takes one to three months of part-time use. Building with APIs takes three to six months. Reaching junior machine learning engineer level typically takes twelve to twenty-four months. Research-adjacent work takes years and usually a postgraduate route.
Do I need to know maths to learn AI?
Not to start. You can use models, build with APIs and do evaluation work without it. You need linear algebra, calculus and statistics to debug training, design loss functions and read papers. Learn it when a specific problem demands it rather than in advance.
Do I need to learn Python?
For anything beyond tool use, yes. Python is the working language of the field and the APIs, libraries and job listings assume it. For evaluation, annotation and words-centred AI work, careful reading and clear writing matter more.
Can I learn AI for free?
Yes. Vendor certification study paths, several full university courses and framework documentation are all free, and the top of that material is better than most paid courses. What money buys is structure, accountability and a credential rather than better content.
What should I build first?
One small thing that solves a real problem, finished and deployed. Something that automates a tedious part of your own job is ideal. Avoid another chatbot wrapper or a dataset that forty other learners used the same week.
Is it too late to learn AI in 2026?
The listings we screen do not suggest so, though the picture is mixed rather than rosy: 59% of currently open listings ask for advanced skills and only 11% are explicitly beginner-friendly. The market is real and it is not undemanding.
Mufy Pachorawala · Founder, getAIwork
My AI scans thousands of AI-job posts (39,308 screened so far) and I personally approve every listing before it reaches the board. I write these articles by the same rules: pay quoted only as listed, and no income promises. Read our editorial rules.
Related: Best AI tools in 2026 · Best AI certifications in 2026 · AI jobs with no experience · AI jobs in 2026
