Back to index
OtherOne video of about 67 minutes, plus a one-page whiteboard summary·Free on YouTube

AI-901 Azure AI Fundamentals Study Cram by John Savill

4.6

The best hour you can spend before sitting AI-901, and it costs nothing. It is a cram, so it organises what you need to know and shows it working, but it will not replace doing the hands-on work the new exam expects.

What We Liked

  • Free, with a downloadable whiteboard that works as a one-page revision sheet
  • Recorded for AI-901 specifically, so it covers Foundry, agents and the SDK instead of recycling AI-900 material
  • Savill shows the portal and real code, including Entra authentication versus keys, which the exam does ask about
  • Chaptered properly, so you can jump back to speech or information extraction the night before
  • He tells you up front which Microsoft Learn resources to pair it with

What Could Be Better

  • 67 minutes cannot give you the Python and SDK practice that makes up more than half the exam
  • No practice questions
  • Foundry changes often, so screens and menu names will drift from what is in the video
  • The pace is quick and assumes you are comfortable with basic Azure ideas like subscriptions and identities

Detailed review

John Savill has been the default answer to how do I pass an Azure exam for years, and I think the reputation is earned. His study crams are free, and the description of this one links to the whiteboard he drew, a GitHub repository of his certification material and a charity T-shirt store. Nothing in it asks you to buy a course. This one covers AI-901, the exam that replaced AI-900 when Microsoft retired it on 30 June 2026. That change matters for prep, because the new exam is over half hands-on work in Microsoft Foundry and most of the older AI-900 videos and question banks are now teaching the wrong test.

Savill published this cram in May 2026 and it had passed 100,000 views when I checked in November. It runs a little under 67 minutes. The first ten cover what AI is, assistants versus agents, and responsible AI. Then he moves into Azure and Entra basics, Foundry itself, choosing and deploying models, building an agent with tools and knowledge, and the difference between authenticating with keys and with Entra. There is a section where he takes the code Foundry generates and builds a small custom client from it, which maps directly to the exam objective about lightweight chat applications.

The second half goes through generative AI, embeddings, natural language processing, speech, vision, image and video generation, and information extraction with Content Understanding. What I like about his teaching is the whiteboard. He draws one picture over the whole session and by the end it is a map of the syllabus. You can download it and use it as your revision sheet. He also explains why things are built the way they are, which makes exam questions easier to reason through when you have not memorised the specific answer. You should be clear about what a cram is, though.

It is a summary and a set of demos. AI-901 expects you to have written prompts, called a deployed model from the SDK and tested an agent, and watching someone else do it for an hour is not the same thing. He says as much and points you to Microsoft Learn. There are also no practice questions here, so use the free official practice assessment for that. My other caution is shelf life.

Foundry has been renamed twice in about two years and the portal moves around. If you are reading this well into 2027, check the exam study guide change log against the video chapters before relying on it. For a $99 exam, I would not pay for a prep course. This video, Microsoft Learn and a few evenings in the portal is the right amount of effort.

[ final ]

The verdict.

Watch it twice: once at the start to get the map, once the day before the exam. In between, do the Microsoft Learn path and the free practice assessment and actually deploy a model and an agent. For a fundamentals exam that is all the preparation most people need, and none of it costs money.