Most people who sign up for an AI class online quit somewhere around week three. Not because the maths is brutal or the instructor is bad. They quit because they never decided what they were building toward, so every module felt like one more thing to memorise instead of one more step toward something they wanted.
This is a plan for the other path. It assumes you have a job, about 45 to 60 minutes a day, and no PhD. Follow it loosely and you’ll have something working by the end of week eight. Follow it exactly and you’ll have a portfolio piece by week six.
Decide the Destination Before You Shop for a Class
Course catalogues are designed to be browsed. That’s the trap. Before you open a single syllabus, write one sentence describing what you want to be able to do in eight weeks. It has to be something someone could watch you do on a screen share.
Bad version: “Understand machine learning.” You can’t demo understanding, so you’ll never feel finished.
Good versions look like this:
- Build a chatbot that answers questions from my company’s onboarding PDFs, and explain how the retrieval step works.
- Automate a weekly report that used to take me four hours, using Python and an API call to a language model.
- Fine-tune a small image classifier that sorts product photos into six categories with better than 85% accuracy.
Notice all three are testable. That matters more than it sounds, because the single biggest predictor of whether you finish an AI class is whether you can tell you’re making progress. If you’re still comparing options, it helps to know what actually separates a useful AI class from an expensive video library before you hand over a card number.
The Eight-Week Skeleton, With Honest Hours
Sixty hours total is enough to go from curious to competent on a narrow task. Here’s how to split it.
Weeks 1–2: Foundations, But Only the Useful Half
You need four ideas before anything else makes sense: training versus inference, what a vector is, why models need data splits, and how a loss function tells you you’re wrong. That’s roughly six hours of reading and video, not forty.
Spend the rest of these two weeks on Python. Loops, dictionaries, list comprehensions, reading a CSV, and calling a REST API with the requests library. If you can do those five things, every AI course online becomes far easier to follow. If you can’t, you’ll spend week five debugging syntax instead of learning models.
Weeks 3–4: Go Deep on One Library, Not Five
Pick scikit-learn if your goal is tabular data and quick wins. Pick PyTorch if you’re heading toward deep learning. Then ignore everything else for two weeks. The person who half-knows five libraries loses to the person who genuinely knows one.
A realistic target for week four: take a messy dataset from Kaggle, clean it, train two models, and write down which one won and why. Two paragraphs of written reasoning is worth more than three notebooks you never reopened.
Weeks 5–6: Build Something Ugly on Purpose
This is where most self-directed learners stall, because they try to build something impressive. Build something small and slightly broken instead. A Streamlit app with one input box and one output. A script that runs on a schedule and emails you a summary. Ugly is fine. Working beats polished.
Give yourself a hard 10-hour ceiling per project. If you blow past it, the scope was wrong, not you.
Weeks 7–8: Make It Explainable
Take the ugly thing from weeks five and six, and do three passes over it. First, delete every line of dead code. Second, write a README that explains what it does in five sentences. Third, record a three-minute walkthrough where you narrate your decisions out loud. That recording is the artefact that gets you a job or a promotion, not the code.
Match the Format to Your Actual Week
Not every format suits every schedule. Here’s a rough map, based on how much structure you need and how much you’re willing to pay for it.
If you’re paying nothing and self-motivated, start with free options. Alison’s free AI courses are a reasonable place to test whether you’ll actually open the laptop on a Tuesday night, and the price tag means quitting costs you nothing but time.
If you need a human watching your screen, live instruction changes the equation. Noble Desktop’s AI training leans on that instructor-in-the-room model, which suits people who stall when left alone with a video queue.
Career switchers tend to want a cohort, deadlines, and a placement conversation. That’s the pitch behind General Assembly’s AI bootcamp, and it’s worth understanding what you’re buying before committing that kind of money.
Then there’s the certification track. If your employer reimburses training or your industry filters on credentials, Simplilearn’s AI program is built for exactly that scenario. It won’t make you a researcher, and it isn’t trying to.
The 25-Minute Rule
Long weekend sessions feel productive and teach you less. You forget 60% of a four-hour block by Wednesday, and the momentum dies with it. Twenty-five minutes, five days a week, beats a Saturday marathon almost every time.
Keep a running text file called failures.md. Every time something breaks for more than ten minutes, write one line about it. Shape mismatches. Missing API keys. A CSV with three different date formats. This file becomes the most valuable document you own, because it’s the only one tailored to your specific blind spots.
What to Do When Week Five Feels Impossible
There’s a predictable wall about five weeks in, right when foundations end and building begins. Your tutorials made sense. Your own project doesn’t. That’s not a signal you lack talent; it’s the normal gap between following and producing.
Three things get you through it. Read someone else’s finished project on GitHub and copy one idea from it, not the whole thing. Rebuild a tutorial you already completed without looking at the video, and note where you get stuck. Explain your bug out loud to an empty room, because saying it forces you to notice the assumption you skipped.
If you’re stuck for three straight days, shrink the problem. Can’t get the model to converge? Hardcode the predictions and finish the app around them. Get the pipeline working end to end first, then improve the parts.
Make the Work Visible
Skills you can’t show are skills nobody pays for. By week eight you should have three public artefacts: a GitHub repository with a README that explains a decision you made and a mistake you corrected, a short written post about something you got wrong and how you fixed it, and that three-minute walkthrough recording.
Update one of them every month after that. Not to impress anyone — to keep the habit alive. The people who stay competent with AI tools aren’t the ones who took the best class. They’re the ones who kept shipping small, unglamorous things long after the course ended.