If you did my first 7 days, this is the part you came for. Kaggle prints a "Builds on" field on every course page, and it is doing real work here. Three of the seven days build directly on courses from that earlier week:
So these are not repeats, they are the next rung. You already did most of the prerequisites without knowing it. The original seven are here if you want to check. One prerequisite to pick up first: day 5 assumes Pandas. It is a free Kaggle course like the rest, so pick it up before day 5 if you have not used Pandas. Kaggle will let you begin any of these cold, but they assume the earlier material and you will feel it by lesson two.
This section matters more on this list than on any of my others, because Google Skills is not a uniformly free platform and I don't want anyone caught out.
The Google Skills days, 1 to 4. Here is what most write-ups get wrong about this, and where the real proof is. Those course pages do not price the course at all, in either direction. Where the word "free" turns up near them it belongs to a Google Cloud free-trial banner advertising cloud credits, which has nothing to do with what the course costs. The real pricing proof is the Google Skills subscriptions page: Starter at no cost with 35 credits a month, Pro at $29 a month.
Where the real proof is. The Google Skills subscriptions page is the page that actually prices the platform: the Starter tier is listed at no cost and comes with 35 free credits a month, and Pro is $29 a month. That is what makes these four days free. It is also what makes "everything on Google Skills is free" untrue, and I'm not going to tell you it is. Credits are spent on hands-on labs, and none of these four days contains one.
The Kaggle days, 5 to 7. No ambiguity at all. All three pages print it in the sidebar: "Cost: No cost, like all Kaggle Learn Courses", and the certificate is included.
Open the course 路 45 minutes 路 Introductory 路 badge, no coding, 9 videos and 4 quizzes
What it is. The right place to start, and the most popular item on this list by a wide margin, with tens of thousands of ratings behind it.
What it actually teaches. What generative AI is, what it is used for, and how it differs from traditional machine learning. That last distinction is the useful one, because it is the thing most people get wrong in conversation.
What you walk away with. A Google Skills badge, and the vocabulary the other three Google days assume you have.
The honest bit. A free Google account is needed to enrol and to earn the badge. Do this one first even if you think you know the material, because days 2 to 4 build on the language it sets up.
Open the course 路 1 hour 路 Introductory 路 badge, no coding
What it is. The models underneath everything else. Day 1 told you what generative AI does; this one is about the thing actually doing it.
What it actually teaches. What large language models are, where they are genuinely useful, and prompt tuning, which is the part most people skip and then wonder why their outputs are inconsistent.
What you walk away with. A Google Skills badge, and a real mental model for why an LLM behaves the way it does.
The honest bit. This is the shortest contents list of the four, just one video, one reading and one quiz, so the hour is mostly the video. It is denser than it looks, and it is worth pausing rather than letting it run.
Open the course 路 30 minutes 路 Introductory 路 badge, no coding
What it is. The pivot point of the week. Days 1 and 2 were about models that respond. This is about systems that act.
What it actually teaches. How agents reason and act, and the architecture behind them: models, tools and orchestration. Those three words are the whole concept, and half an hour is genuinely enough to make them stick.
What you walk away with. A Google Skills badge, and the ability to follow any agent conversation without nodding along.
The honest bit. This is the oldest page of the four, last updated around five months ago, so it predates some of this year's agent tooling. The architecture it teaches has not changed, which is why it still earns its place.
Open the course 路 1 hour 30 minutes 路 Introductory 路 badge, no coding
What it is. The longest of the four Google days and the freshest page of the set, updated within the last month. It goes a layer deeper than day 3.
What it actually teaches. Where agents add real value, using autonomous goal-directed behaviour as the lens. Less "what is an agent" and more "when is an agent the right answer", which is the more useful question and the one almost nobody covers.
What you walk away with. A Google Skills badge, and a way of judging agent proposals rather than just understanding them.
The honest bit. Google describes the audience as developers, architects and technical decision-makers. There is still no coding in it, but the framing assumes you are the person deciding whether to build something. If that is not you yet, it is still worth doing; just expect the examples to be organisational rather than personal.
Open the course 路 7 lessons 路 4 hours to earn the certificate 路 free certificate, Python in the browser
What it is. The switch from watching to doing, and the first of the three Kaggle days. Everything before this was concepts. From here you are running code.
What it actually teaches. The parts of a real model that the introductory course leaves out: handling missing values and categorical data, building pipelines so your steps stay in order, cross-validation, gradient boosting with XGBoost, and spotting data leakage before it quietly ruins your results. Leakage is the one worth the whole day, because it is the mistake that makes a model look brilliant right up until it meets real data.
What you walk away with. A Kaggle certificate of completion, and a model-building process you can actually repeat rather than a notebook you followed along with.
The honest bit. Kaggle prints the cost plainly: "No cost, like all Kaggle Learn Courses". It lists two things it builds on: Intro to Machine Learning, which was day 1 of my first seven, and Pandas, which was not on that list at all. If you have never used Pandas, do that free Kaggle course first or this one will feel harder than it is. All the code runs in the browser in Kaggle Notebooks, so there is nothing to install.
Open the course 路 5 lessons 路 4 hours (estimated) 路 free certificate, Python required
What it is. The step up, and the first of the three Kaggle days. Taught by Dan Becker. This is where the week stops being conceptual.
What it actually teaches. Permutation importance, partial dependence plots and SHAP values. In plain terms: how to work out why a model decided something, rather than only what it decided. That is the skill that turns a model from a black box into something you can defend in a meeting.
What you walk away with. A Kaggle certificate of completion, downloadable and shareable to LinkedIn, plus the most job-relevant technical skill on this entire list.
The honest bit. Kaggle's page says "Builds on Intro to Machine Learning", which was day 1 of my first seven, so if you did that list you are ready. There is real Python here, though it runs entirely in the browser in Kaggle Notebooks with most of the code written for you. Kaggle words the duration as "4 (estimated)" and that is fair; it depends on how much you experiment.
Open the course 路 6 lessons 路 4 hours (estimated) 路 free certificate, Python required
What it is. The finale, and the most satisfying thing on the list, because you finish the week having built something that works rather than having understood something. Taught by Ryan Holbrook.
What it actually teaches. Building a convolutional image classifier with TensorFlow and Keras, from the convolution operation through to a trained model that recognises images.
What you walk away with. A Kaggle certificate of completion, and a working image classifier you built yourself, which is a far better interview answer than a badge.
The honest bit. It says "Builds on Intro to Deep Learning", which was day 6 of my first seven. This is the steepest item in the week, and it is the one to move to a Sunday rather than squeeze into a weeknight. All the code runs in the browser, so there is nothing to install.