Drafting a one-page AI use-case memo by hand runs about three hours of reading, writing and trimming. After Anthropic's free AI Fluency course it takes 45 minutes, review included, and that gap is the whole point of this list.
Anthropic put 10 free AI courses on Coursera on October 4, 2026, taking its catalog from 12 to 22. The shortest runs 2 hours. Set that against a bootcamp price and a headline number of $5,000 a month.
$5,000 a month is a goal, not a guarantee. A University of Oxford study of 1,600 recruiters in the US, UK and Germany found certified AI skills made applicants more likely to be shortlisted for interviews. It found nothing about pay.
The certificate is the weakest thing a course gives you. The artifact you build, whether a memo or a working chatbot, is what a client or hiring manager pays for.
Pick the course that fits your situation, and finish one piece of proof before spending a cent.
| Course | Provider | Time to finish | Free access | Certificate or exam cost | Best for |
|---|---|---|---|---|---|
| AI Fluency: Framework & Foundations | Anthropic on Coursera | 3 hours | Yes | Free to learn | Non-coders who judge AI use cases |
| AI Capabilities and Limitations | Anthropic on Coursera | 4 hours | Yes | Free to learn | Anyone reviewing AI output |
| Claude Code 101 | Anthropic on Coursera | 2 hours | Yes | Free to learn | Developers directing coding agents |
| ChatGPT Prompt Engineering for Developers | DeepLearning.AI | Not stated, self-paced | Yes | See course page | Developers building a chatbot |
| Google Generative AI Leader | 7 to 8 hours | Yes | $99 certificate, 90-minute exam | Managers running adoption | |
| Introduction to AI in Azure | Microsoft | Just over 10 hours, 14 modules | Yes | See Microsoft Learn | IT staff in Microsoft shops |
| AWS Fundamentals of ML and AI | AWS | About 1 hour, 1 module | Yes | See AWS page | Cloud beginners needing vocabulary |
| Google Cloud ML and AI path | Google Cloud | Self-paced path | Yes | $200 exam | Engineers aiming at cloud ML roles |
| Practical Deep Learning for Coders | fast.ai | Self-paced lessons | Yes | Free | Developers who learn by shipping |
| Hugging Face LLM Course | Hugging Face | Self-paced | Yes | Free certificate | Python developers fine-tuning models |
1. AI Fluency: Framework & Foundations

Three hours of vocabulary and judgment will not hand anyone a coding portfolio. Treat this as the groundwork you lay before a build course, not a substitute for one. It is also the starting point Anthropic itself recommends before its role-specific material.
The launch behind it: 10 free courses on Coursera from October 4, 2026. This one runs 3 hours, beginner level, no code.
What it leaves in your hands is the shared language for arguing about AI at work, which is what managers, teachers, owners and creators actually need.
By hand, writing an AI use-case memo for a team means a day of reading, notes and redrafts, often without a shared vocabulary.
Here the lessons come in at about 3 hours, the draft memo takes a chat assistant under a minute, and your edit pass runs 25 to 35 minutes.
The split: 3 hours of lessons plus 30 minutes of editing against most of a working day from scratch. The memo is the artifact. Brief colleagues on where AI fits and where it does not, and avoid overclaiming.
💡 Tip: Take AI Fluency before any other Anthropic course. The role-specific ones assume its vocabulary.
2. How to Tell When an AI Answer Is Wrong

Shared AI vocabulary still leaves the hard question open, which is whether a confident answer is right. This four-hour beginner course on Coursera trains the checking habit instead.
It covers model evaluation and AI literacy, free on Coursera and aimed at the beginner who has just learned the terms. Analysts, editors and operations staff who review AI output before it goes out are the natural fit.
By hand, a careful editor checks every fact and link in a 700-word AI-drafted summary. That is 45 minutes with three browser tabs open. With the course checklist it is faster, because the questions are fixed.
The catch is that every major assistant can state false facts with confidence and invent sources, quotes and links. This course is not a trust shortcut. It is training in catching the errors, not a license to skip the check.
The portfolio proof is a five-question eval checklist applied to one task on your desk, such as summarizing a report. Five checks over 700 words take about 15 minutes and catch most invented names and links.
Draft first, then run the checklist. That order makes the errors visible instead of letting a polished summary pass review and reach a client.
3. Claude Code 101

Reading a diff line by line, then retyping the fix into a file, is the dullest part of coding. Judging an AI answer is the skill that comes before trusting an agent to make those edits for you.
Claude Code 101 runs two hours, beginner level, free on Coursera, on coding agents and context engineering. An agent here means a tool that edits files on your machine when you describe the change. Its folder is your repository, the version-controlled home of the project.
Keep that repository a sandbox copy with version control, never a live client project. The check is one command that shows the agent's proposed change before it touches anything. Most beginners skip it.
The course is free, and the real cost is hours. Two hours of lessons, plus about two hours outside class to build the small command-line tool and push it to a public repo. That is four hours, $0 for the course.
Total: 2 hours of lessons, about 2 hours of practice, $0. What you bring back is a loop for briefing, reviewing and testing agent-written code.
Course words, translated
4. ChatGPT Prompt Engineering for Developers

Yes, and in about the time a short course takes. DeepLearning.AI's ChatGPT Prompt Engineering for Developers is free and self-paced, and it uses an API to build a custom chatbot.
An API is the doorway that lets your own code talk to an AI model directly.
That doorway is the whole point. Chatting in a browser tab is not the same as shipping a prototype your code controls. Before you paste anything into a chatbot, know the privacy rule: never paste passwords, client files, or medical or financial records.
The research states no hours for this course, so check its page for current requirements before blocking a weekend.
Keep the bot on one narrow task, such as answering returns questions from your own policy text, and demo that. A working prototype outranks a certificate in a client conversation.
💡 Tip: Scope the chatbot to one narrow task. A bot that answers questions about a single policy doc demos better than a general assistant.
💡 PRO TIP
Build the proof before the certificate
Every course here leaves you with something to show. Finish that artifact first, then decide whether any certificate is worth paying for. Recruiters ask what you built, not which badge you collected.
5. Google Generative AI Leader

A one-page adoption brief, about 7 to 8 hours after you start. Building a prototype proves you can make one. This course is for whoever decides whether a team should make one at all, and Google's Generative AI Leader maps generative AI to business decisions.
Step one: work through the free learning material and list three tasks your team does by hand.
Step two: name one where AI fits and one where it does not, with the reason for each. You should end with a two-column list you can defend.
Step three: add a time cost to the task you keep, then write the brief. Learning is free, but the optional certificate costs $99 and the 90-minute exam, so skip it unless an employer asks.
Walk into a budget meeting able to argue the case, having paid nothing for the knowledge itself.
6. Introduction to AI in Azure

Prompt to copy
You are helping me write a service-selection note for my IT team. Our company already runs Microsoft Azure. Below are three business tasks we do manually each week: [task 1], [task 2], [task 3]. For each task, name one Azure AI service that fits, say in one sentence what the service does, and give one reason it fits better than doing the task by hand. Output a three-row table with the columns Task, Azure Service, Reason. Keep each row under 30 words. Do not recommend services outside Azure.
Paste that into ChatGPT or Claude once you know your three tasks. What comes back is a mapping you can check line by line against Microsoft's own service descriptions, and that check is the work. The catch is scope: this course is Azure-specific.
It is not a neutral survey of every AI platform, and it will not transfer as one. Microsoft's Introduction to AI in Azure is free and self-paced, spans 14 modules and takes just over 10 hours, so block it across two or three evenings.
The payoff is a service-selection note that goes straight into a live Azure project, written in the vocabulary your admins already use.
What the free track really costs
Prices as of November 2026. Confirm on each provider's own page.
7. AWS Fundamentals of Machine Learning and AI

One hour buys you orientation, not a job. That is the honest size of AWS Fundamentals of Machine Learning and Artificial Intelligence: a single module, free and self-paced, done in about the time of a lunch break and a coffee.
Walk in expecting a map, not a skill.
It suits cloud beginners who need the vocabulary before they commit to a longer AWS path. You meet the terms and the service names once, in plain context.
Then you write your own one-page glossary and a shortlist of three AWS services to explore next. That page is small, but it is a real artifact you can hand someone.
The catch is scale. Sixty minutes cannot make anyone job-ready, and the course alone will not hand you a portfolio piece a client would pay for. The check is simple: read a junior AWS AI job posting cold, and underline every term you cannot define.
If you clear that read, the course did its job. The payoff is smaller than a certificate and more useful: you can open an AWS AI job description or an architecture diagram and keep moving, instead of stopping at every unfamiliar term.
💡 Tip: Pair it with a hands-on AWS lab the same week. Vocabulary fades fast without one click in the console.
Certificate or portfolio: the same $5,000 goal, two routes
Paying for the certificate
- $99 for Google's 90-minute exam
- $200 for the Google Cloud ML exam
- Credential appears on a profile
- Proves completion, not capability
- Shortlists you, then the interview starts
Building the proof instead
- A working chatbot you can demo live
- A documented fine-tuning notebook
- $0 spent beyond metered API calls
- Shows what you can do, not what you finished
- Opens the conversation on your terms
8. Google Cloud Machine Learning and AI Path

Yes, but not a free one.
The path itself is free to access, while the Professional Machine Learning Engineer exam costs $200. That is the deepest of the three cloud options in this list.
It fits engineers aiming at cloud ML roles, not admins supporting them. The portfolio piece is a one-page design doc for a system you could actually build, covering data, model, serving and evaluation.
Write it before you book anything, and the exam becomes a business decision instead of a leap.
Use the path and the design doc to test a single question: is this exam worth $200 for the job you are actually targeting? If the posting you want names the credential, fine. If it names shipped systems, put the money back in your pocket.
9. fast.ai Practical Deep Learning for Coders

Reaching a working image model takes one lesson here. Practical Deep Learning for Coders by fast.ai is free and teaches from the top down, so the first session ends with a trained model doing real classification, not a page of theory.
That approach has a cost. You will be running code before you fully understand what sits under it, so plan review time after each lesson. This course is for developers who learn by shipping, not by finishing the math first.
There is a limit on what AI can do for you at this depth, and it is worth saying plainly. A chatbot can explain the code, but it cannot give you the understanding of model internals that an interview question tests. Do the runs yourself.
The artifact is a working classifier you can deploy or at least run from a repo. Open it, explain the choices, and that demo carries more weight than a completion badge.
🚀 From zero to a model you can demo
Take AI Fluency
Three hours to get the shared vocabulary before touching any code.
Add the limitations course
Four hours of eval training so you can check what a model tells you.
Run Claude Code 101
Two hours on directing a coding agent inside a sandbox repo.
Start fast.ai lesson 1
Train an image model the same week, before the theory catches up.
Deploy the classifier
Put the model somewhere a stranger can open it on a phone.
Fine-tune with Hugging Face
Take the documented notebook to interviews as your deepest proof.
10. Hugging Face LLM Course

Prompt to copy
I am taking the Hugging Face LLM Course and want a portfolio project, not a certificate to hang on the wall. My setup: [your laptop, whether you have a GPU or just a Colab notebook, your Python comfort level]. Course covers tokenizers, the transformers library and fine-tuning. Give me one narrow fine-tuning task I can finish this weekend (a classifier, not a chatbot), the exact notebook cells in order, the dataset to load, the 3 training settings I should start with, and the one metric to record before and after. Keep it under 400 words, skip the theory, and flag anything that will need more compute than a free Colab session.
That prompt goes into Claude or ChatGPT and asks for a fine-tuning project sized to a weekend. A language model is the model itself, a system trained on huge amounts of text to predict what comes next.
Fine-tuning trains a copy of a small one on your own labeled examples, so it gets good at one narrow job and nothing else.
What comes back is a plan: a dataset, a notebook skeleton, and the three settings to start with. Review it in about 20 minutes, because the settings are where it goes wrong.
If the loss figure stalls or the model gets worse after training, cut the learning rate. The certificate says completion; the documented notebook is the proof.
Frequently Asked Questions
Do I need to know how to code to start any of these?
No. AI Fluency, AI Capabilities and Limitations, and Google Generative AI Leader are all beginner level with no coding. Claude Code 101 assumes you can read code, and fast.ai expects Python comfort by lesson 2.
Is $5,000 a month realistic from free courses alone?
The courses build the skill and the proof; finding buyers is separate work. Treat $5,000 as a target that depends on clients or a role, not a figure any course promises. The Oxford recruiter study suggests certification helps you get shortlisted, not that it sets your rate.
Which course should I take first if I only have three hours?
AI Fluency: Framework & Foundations. It is three hours, free on Coursera, and Anthropic recommends it before its role-specific courses. You finish with a use-case memo you can show someone.
Are these really free, or is there a catch?
The learning is free. Two things cost money and both are optional: Google's Generative AI Leader certificate at $99 and the Google Cloud Professional Machine Learning Engineer exam at $200. API access for the chatbot course is metered separately.
Can I put AI course certificates on my resume?
Yes, and the Oxford study of 1,600 recruiters found certified AI skills were more likely to get applicants shortlisted. Pair the certificate with the artifact the course produced, because the interview will ask what you built.
What should I never paste into a chatbot while learning?
Passwords, client files, and medical or financial records. On free and personal plans, chats may be used to improve the company's models unless that is switched off in settings.
Start with the course that matches what you already do
Total beginner or non-coder: AI Fluency, then AI Capabilities and Limitations. Manager: Google Generative AI Leader. Developer after a fast win: Claude Code 101 or ChatGPT Prompt Engineering for Developers.
Cloud track: Introduction to AI in Azure, then Google Cloud's ML and AI path. Deep ML: fast.ai, then Hugging Face.
The first action is one course, not five. Pick the one on that list, block its stated hours in your calendar, and build the portfolio proof before you pay for any certificate.