AI-901: how Azure AI Fundamentals changes in 2026

Written by Mario Larrúa Vega.

If you saved an AI-900 course to start this month, it is time to review your plan: Microsoft retired that exam on June 30, 2026. The date appears in its official retired exams register.

The certification Microsoft Certified: Azure AI Fundamentals continues, and its current page requires passing AI-901. This distinction matters: the code identifies the exam; Azure AI Fundamentals is the credential you earn. You can check both in the official certification page.

For people starting with Azure, this change deserves attention. The new preparation includes more work with models, applications and agents. That means reviewing what you study and, especially, how you practice it.

Retirement dates, requirements and objectives come from Microsoft. The study path and suggested exercise are LarrúaTech editorial recommendations.

What changes between AI-900 and AI-901

The last AI-900 guide organized content into five areas: AI workloads, machine learning, vision, language processing and generative AI. Its candidate profile included technical and nontechnical backgrounds and did not require software engineering experience. That description remains useful for understanding the previous exam’s starting point. You can consult the historical AI-900 guide.

AI-901 defines two main areas:

Its objectives include deploying a model, working with prompts, creating a lightweight SDK client, testing an agent and developing small text, speech, vision and information-extraction solutions. The reference for exam preparation is the AI-901 study guide.

The implication for your study is clear: set aside time to build. After learning a concept, find a small task that requires you to apply it and explain the result.

The percentages describe knowledge areas. Do not interpret them as a guaranteed breakdown of theoretical questions, coding exercises or labs during the exam.

Why Microsoft Foundry and agents feature in this evolution

Microsoft describes Foundry as a platform bringing together models, agents and tools, with evaluation, observability and access-management capabilities. It also combines portal exploration with SDK development. Its documentation covers the evolution from earlier names, such as Azure AI Studio and Azure AI Foundry, to Microsoft Foundry. This helps when a tutorial uses another name. The explanation is in what Microsoft Foundry is.

An example helps show why this is useful to learn. Imagine an application that handles support questions for a training provider:

• A student asks why they cannot access a lab.
• The application must interpret the problem and consult relevant information.
• Its answer must follow the service’s instructions and acknowledge missing information.
• The team needs to verify whether answers help resolve questions.

Each decision raises a technical question: which model to use, what context to provide, how to verify behavior and what access the application needs.

The official introduction to generative AI and agents in Azure module specifically proposes working with model selection, deployment, instructions, tools and tests in the experimentation environment, or playground.

My interpretation of this change is that a good introduction should help connect concepts with implementation decisions. For a beginner, a small application understood well can teach much more than a collection of demonstrations they have only watched.

What you need to know before starting

The AI-901 page calls for conceptual AI knowledge, basic technical skills, familiarity with Azure resources and knowledge of programming syntax and techniques in Python. It also mentions familiarity with REST APIs, SDKs and command-line tools. This is the profile Microsoft publishes on the exam page.

For practical preparation, we suggest being able to do these tasks before attempting the exercises:

• Read a small program with variables, lists, dictionaries and functions.
• Recognize what data a request sends and what a response returns.
• Understand a simple JSON object.
• Locate the configuration an application uses to connect to a service.
• Explain the difference between creating a resource and having permission to use it.

This list is LarrúaTech study guidance; it does not add official requirements.

The course AI-901T00: Introduction to AI in Azure also recommends knowledge of cloud storage, compute, authentication and authorization. If you have studied AZ-900, you can use that foundation to understand where your solution runs and how it accesses resources.

If you do not code yet, spend a few sessions on Python first. Try changing a small example and explaining each change. That practice will help you get more from labs than copying instructions without understanding them.

If you passed AI-900, your certification still counts

Exam retirement can raise questions about a credential already earned. Microsoft states that Fundamentals certifications do not expire and do not require renewal. The policy can be found in its renewal frequently asked questions.

The retirement documentation also states that earned credentials remain in Microsoft Learn history. You can review that explanation in what happens when an exam retires.

Therefore, if you already earned Azure AI Fundamentals through AI-900, you do not need to pass AI-901 to prevent expiration. You can study the new content to update your skills and decide afterward whether to take the exam.

If you started studying but have not taken the exam, use your notes as a foundation and compare them with the current objectives. Keep what still fits and prepare the areas you are missing.

A study path from theory to practice

This proposal focuses on progressing through outcomes. Adjust its duration to your experience and available time.

1. Organize the concepts

Start with AI concepts for developers and technology professionals, a Microsoft Learn path covering generative AI, agents, language, speech, vision, information extraction and retrieval-augmented generation, known as RAG.

After each topic, write a use case in your own words. For example: extract fields from an invoice, transcribe a voice note or generate a draft answer. Include the expected result and how you would check it is correct.

2. Work in Foundry with a small goal

Continue with Get started with AI applications and agents on Azure. This path includes applications, agents and different AI capabilities; it also covers building a lightweight text-analysis client in Python.

Keep a simple record of each exercise: what you wanted to achieve, what you configured, what result you obtained and what you would change. When something fails, record the cause you have verified.

3. Check answer quality

Prepare several cases before changing your application’s instructions. Include a clear question, an ambiguous question and one that cannot be answered with the available information.

Use the same cases after each change. This helps detect improvements and regressions without relying on a single answer seeming convincing.

4. Check your gaps against the syllabus

Mark each guide objective as “I can explain it”, “I have practiced it” or “I need to review it”. That classification gives you a specific next task.

A result such as “the client works, but I do not understand where it gets its configuration” is useful: it identifies exactly what to investigate before continuing.

Suggested lab: a support assistant for a training provider

Microsoft offers a guided exercise on generative models and agents in Foundry. It lets you deploy and explore a model, then use it in an agent with knowledge tools. You need an Azure subscription to complete it independently.

After completing that exercise, you can adapt it to the following fictional scenario. It is a LarrúaTech practice suggestion, independent of the exam.

Goal: guide students who have trouble accessing a lab, using only a short help guide you prepare.

Write a guide with three fictional incidents, initial checks and the circumstances requiring human intervention. Use fictional data and review resource cost and availability before deploying.

These could be initial instructions to experiment with:

Eres un asistente de soporte de una academia de formación.

Utiliza la guía de ayuda proporcionada para orientar al alumno.
Si faltan datos para identificar el problema, pide una aclaración.
Si la guía no contiene una solución, indícalo y propone escalar el caso.
No afirmes que has cambiado permisos ni resuelto una incidencia.

Responde con: problema entendido, siguiente comprobación y motivo.

Check its behavior using a table like this:

Instructions must be accompanied by permissions and tools appropriate to the application’s scope. This exercise only requires consulting information and suggesting checks.

Save the results, change one instruction and repeat the cases. Then try explaining what improved and what still fails. The value lies in that comparison; completing this exercise alone does not cover the entire syllabus.

What to review before booking AI-901

The current page includes Spanish among the exam languages and links to a practice assessment in AI Skills Navigator, which requires signing in. Also review the price for your country or region when booking. Find current links on the official AI-901 page.

The minimum passing score is 700. Microsoft uses a scaled score, so this number does not necessarily equal answering 70% of questions correctly. This is explained in its exam scoring documentation.

Before choosing a date, check that you can justify your exercise decisions: why you used a particular capability, what information the application needs and how you evaluated the result. Use the practice assessment to find gaps and revisit the relevant topics.

Your next step

Open the AI-901 guide and choose an objective you have not practiced yet. Turn it into a small task, run several tests and document what you learn.

That record can become a portfolio entry: the problem, the solution you tried, results and limitations. While preparing for certification, you will build concrete examples to explain your skills.