If you studied for AI-900, don't assume AI-901 is the same exam with a new number. AI-901 is the current exam for Microsoft Certified: Azure AI Fundamentals, AI-900 was retired on June 30, 2026, and the new exam adds hands-on work with Microsoft Foundry. This guide follows the skills outline dated April 15, 2026.
Exam at a glance
| Domain | Weight |
|---|---|
| Identify AI concepts and capabilities | 40-45% |
| Implement AI solutions by using Microsoft Foundry | 55-60% |
Microsoft expects conceptual AI knowledge, basic Python syntax and familiarity with Azure resources. You won't write large programs, but you should be able to read a short client script and know what it does.
Domain 1: AI concepts (40-45%)
Responsible AI is six principles. Memorize them as a set, because scenario questions ask you to match a situation to one:
- Fairness
- Reliability and safety
- Privacy and security
- Inclusiveness
- Transparency
- Accountability
Model components and configuration: how generative models work at a high level, choosing a model by capability, and knowing deployment options and parameters (temperature and similar settings).
AI workloads: generative and agentic AI, text analysis (keyword extraction, entity detection, sentiment, summarization), speech recognition and synthesis, computer vision and image generation, and information extraction from text, images, audio and video.
Domain 2: Implementing with Foundry (55-60%)
This is the larger half, and it's the main difference from AI-900.
- Generative AI apps and agents: write effective system and user prompts, deploy a model and interact with it in the Foundry portal, build a lightweight chat client with the Foundry SDK, and create and test a single agent.
- Text and speech: a lightweight app with text analysis, responding to spoken prompts with a multimodal model, and Azure Speech in Foundry Tools.
- Vision and image generation: interpreting visual input in prompts, generating images, and building a small vision app.
- Information extraction: pulling data from documents, images, audio and video with Azure Content Understanding in Foundry Tools.
Prompting is testable
A system prompt sets behavior; a user prompt carries the request. A weak system prompt leaves the model guessing:
You are a helpful assistant.
A stronger one defines role, scope, format and limits:
You are a support assistant for a software company.
Answer only questions about our product.
Reply in three sentences or fewer.
If you don't know, say so and suggest contacting support.
Expect questions that ask which change improves reliability or keeps the model on topic.
Hands-on checklist
Rather than reading about Foundry, do these once:
- Create a Foundry project and deploy a chat model.
- Chat with it in the portal and change the system prompt.
- Adjust a parameter and compare outputs.
- Create a simple agent and test it.
- Run a document through Content Understanding and read the extracted fields.
Common traps
- Confusing a workload (what you want done) with a service or model (what does it).
- Mixing up transparency (people understand how the system works) and accountability (people are responsible for it).
- Treating this as a vocabulary-only exam. The Foundry half expects you to know the steps.
Study plan (about 3-4 weeks part-time)
- Read the official skills outline and mark gaps.
- Complete the Microsoft Learn material for the exam.
- Do the hands-on checklist above.
- Take the free practice assessment on AI Skills Navigator and review mistakes.
- Try the exam sandbox.
If you already hold the Azure AI Fundamentals credential, Microsoft says no action is needed because the certification itself wasn't retired.
Check the official AI-901 study guide before booking, since skills lists get revised.
What's the first thing you'd build in Foundry to prepare? Tell me in the comments.
Top comments (0)