AI-900 vs AI-901: what changed in Azure AI Fundamentals

The short version

AI-900 was a describe-and-identify exam. Every one of its five skill areas began with "Describe", and its audience profile said data science and software engineering experience were not required.

AI-901 has two skill areas. The first, identifying AI concepts and capabilities, is close to the old exam. The second, and larger, is implementing AI solutions by using Microsoft Foundry: deploying models, writing prompts, and building lightweight chat, agent, speech, vision and information-extraction applications.

The biggest change: you now need some Python

Microsoft’s AI-901 audience profile says you need "knowledge of Python coding syntax and programming techniques" and should be familiar with Azure resources. AI-900 had no coding expectation at all.

The skills outline shows why. You are expected to create a lightweight chat client with the Foundry SDK, create a lightweight client application for an agent, and build small applications for text analysis, Azure Speech, vision and Content Understanding. Reading and reasoning about short Python snippets is part of the exam now, not an optional extra.

What carried over from AI-900

If you studied for AI-900, a good part of that knowledge still counts. These topics appear in both outlines:

• The six responsible AI principles, worded identically: fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability.
• Text analysis techniques: key phrase or keyword extraction, entity detection, sentiment analysis.
• Speech recognition and speech synthesis.
• Computer vision workloads and generative AI scenarios.
• Extracting information from documents and forms.

What is new in AI-901

These have no real equivalent in the AI-900 outline:

• Agents: creating and testing a single-agent solution in the Foundry portal, and a client application for it.
• Model choice and deployment: identifying an appropriate model based on its capabilities, and appropriate deployment options and configuration parameters.
• Prompting: creating effective system and user prompts for generative AI models.
• Multimodal prompting: responding to spoken prompts and interpreting visual input with a deployed multimodal model.
• Azure Content Understanding: extracting information from images, audio and video as well as documents.

What dropped out

AI-900 had a whole skill area on the fundamental principles of machine learning: regression, classification and clustering scenarios, features and labels, training and validation data, and Azure Machine Learning capabilities such as automated machine learning. None of that is in the AI-901 outline.

AI-900 also named specific services to describe, including the Azure AI Face detection service and translation as an NLP workload. AI-901 frames the same ground around building with Foundry rather than describing individual services.

If you were preparing for AI-900

Keep your notes on responsible AI and the workload types; they transfer directly. Drop the classic machine-learning material. Then spend most of your remaining time on the Foundry half of the outline, because it is the larger share of the exam and the part AI-900 never tested.

Be careful with old AI-900 practice questions. Many test the machine-learning area that no longer exists, or name services by titles that have since changed, so a high score on them says little about AI-901 readiness.

Plan your prep

Questions, answered

Sources

Exam facts in this guide come from the vendor's published exam guide. Vendors revise these — check the source for the current version before you book.

• Microsoft AI-901 study guide — Microsoft Learn
• Microsoft AZ-900 study guide — Microsoft Learn