Examinotion
Study Guides

AI-901 Microsoft Foundry Domain: Hands-On Study Guide for 2026

The Microsoft Foundry domain is 55-60% of AI-901. This 2026 study guide covers all 15 hands-on skills, the Foundry SDK code you need to recognise, the retired services to ignore and a one-week lab plan.

ET

Examinotion Team

18 min read4 November 2026Updated: 9 November 2026
Abstract 3D blue modules in four groups on a slate platform with green checkpoints, AI-901 Foundry

Last updated: November 2026. Written and fact-checked by the Examinotion team against the official Microsoft Learn study guide, exam page and Microsoft Foundry documentation for AI-901.

TL;DR The Microsoft Foundry domain is worth 55-60% of the AI-901 exam, and every one of its 15 skills is something you build or do. To pass, deploy a model, write a small Python chat client with the Foundry SDK, create and call a single agent, then practise text, speech, vision and Content Understanding labs.

If you are preparing for AI-901: Microsoft Azure AI Fundamentals, the second domain, "Implement AI solutions by using Microsoft Foundry", decides your result more than anything else on the exam [1]. It is the larger of the two domains and it is the part of AI-901 that catches out candidates who expect a reading-only fundamentals exam. This study guide walks through every skill in the Foundry domain, shows the small amount of code you need to recognise, flags the out-of-date material to ignore, and finishes with a one-week hands-on lab plan.

Why the Foundry domain decides AI-901

The AI-901 exam has two domains, and "Implement AI solutions by using Microsoft Foundry" carries 55-60% of the marks. The other domain, "Identify AI concepts and capabilities", carries 40-45% [1][2]. Microsoft's study guide lists the skills measured "as of April 15, 2026", and no newer version of the outline has been announced [1].

AI-901 domain Weight What it tests
Identify AI concepts and capabilities 40-45% Responsible AI principles, model components and configuration, AI workloads
Implement AI solutions by using Microsoft Foundry 55-60% Building small generative AI, agent, text, speech, vision and information-extraction solutions in Microsoft Foundry

The Foundry domain is practical rather than descriptive. Its 15 skills use verbs such as "deploy", "create", "build" and "extract", and six of them ask you to create a "lightweight application" or "lightweight client application" [1]. Microsoft's AI-901 exam page also states that candidates "need knowledge of Python coding syntax and programming techniques" and "should be familiar with REST APIs, SDKs, and CLIs" [2]. AI-901 is now the only exam required for the Microsoft Certified: Azure AI Fundamentals credential [19], and fundamentals certifications do not expire [20]. The coding expectation is a significant change from AI-900, which this exam replaced; our AI-900 vs AI-901 comparison covers the full transition.

The passing score is still 700 on a scaled score. Microsoft confirms that "a score of 700 or greater is required to pass" AI-901 [1]. A scaled score of 700 is not the same as 70% of questions correct, which our guide to what 700 out of 1000 really means explains.

What you need before you start the Foundry labs

Hands-on AI-901 preparation needs three things: an Azure subscription, a Microsoft Foundry project, and basic Python. The official learning path for this domain lists its prerequisite as "a basic understanding of computing concepts and Python" [3].

  • An Azure subscription. Microsoft's Foundry pricing page says "The Foundry platform is free to use and explore", but individual features "are billed at their normal rates", and you need an Azure subscription before you can build agents [22]. New Azure customers can sign up for a free account that includes credit to spend within 30 days [21].
  • A Foundry project. Microsoft Foundry "unifies agents, models, and tools under a single management grouping", and current guidance centres on Foundry projects in the new Foundry portal. Older hub-based projects are accessible in the "Foundry (classic)" portal, while new investment is focused on Foundry projects in the new portal [5].
  • Python and the Azure CLI. The Foundry SDK quickstart installs azure-ai-projects (version 2.3.0 or later) and authenticates with DefaultAzureCredential from azure-identity, after you sign in with az login [8].

Free tiers help you keep costs down. Microsoft says many Foundry Tools have a free tier that you select with the F0 SKU [16]. Delete resources you no longer need after each lab, because model calls and other features are billed at their normal rates [22].

AI-901 uses the current Foundry vocabulary, so old tutorials can mislead you. Microsoft's own "evolution of Foundry" table records that "Azure AI Studio / Azure AI Foundry" is now Microsoft Foundry, "Azure AI Services" are now Foundry Tools, and the Assistants API has been replaced by the Responses API [5]. Our article on the Azure AI Foundry rename covers the naming changes in more depth.

Skill group 1: generative AI apps and agents in Foundry

The first AI-901 Foundry skill group has five skills, and together they form the core of the hands-on domain. Microsoft lists them as [1]:

  1. Create effective system and user prompts for generative AI models
  2. Deploy a model and interact with it in the Foundry portal
  3. Create a lightweight chat client application by using the Foundry SDK
  4. Create and test a single-agent solution in the Foundry portal
  5. Create a lightweight client application for an agent

Prompts and model deployment

For AI-901, model deployment starts in the Foundry model catalogue and ends in the playground. When you deploy a model in the Foundry portal, "you immediately land on its playground", where you can test system and user prompts before writing any code [7]. Practise writing a system prompt that sets the model's role, tone and constraints, then compare how different user prompts change the output.

Deployment types control where data is processed and how you pay. Microsoft's deployment-types page advises: "For most workloads, start with Global Standard", which is billed pay-per-token and processes data in any Azure region [6]. Data Zone deployments keep processing within a Microsoft-defined zone such as the US or EU, and Batch deployments process large asynchronous jobs at 50% less cost than Global Standard, with a 24-hour target turnaround [6]. At fundamentals level, know what Global Standard means and why an organisation might choose a Data Zone option for data residency.

Building a chat client with the Foundry SDK

The AI-901 chat-client skill uses the Foundry SDK to get an OpenAI-compatible client from your project, then calls the Responses API. The Microsoft Learn module for this skill group lists "Deploy a model in the Foundry portal and replicate Playground settings in code" as a learning objective [4]. The current quickstart pattern looks like this [8]:

from azure.identity import DefaultAzureCredential
from azure.ai.projects import AIProjectClient

project = AIProjectClient(
    endpoint="https://{resource-name}.services.ai.azure.com/api/projects/{project-name}",
    credential=DefaultAzureCredential(),
)
openai = project.get_openai_client()

response = openai.responses.create(
    model="gpt-5-mini",  # your deployed model name
    input="Explain responsible AI in one sentence.",
)
print(response.output_text)

You do not need to memorise this code, but you should recognise every part of it. Be ready to identify the project endpoint, the credential, the client and the call that returns the model's answer. Microsoft warns that the quickstart code "uses Azure AI Projects 2.x and is incompatible with Azure AI Projects 1.x" [8], so tutorials written for the 1.x SDK will not run as written.

Creating and calling a single agent

A single-agent solution in Foundry is usually a prompt agent: a model plus instructions, and optionally tools, defined through configuration. Microsoft's Foundry Agent Service overview describes a portal-first path: "create an agent interactively in the Foundry portal, test it in the playground, then call it from your application code" [9]. The agents playground "lets you explore, prototype, and test agents without running any code", including multi-turn conversations [7].

Calling an agent from code adds a conversation that keeps history across turns. In the prompt-agent quickstart, the client gets an OpenAI client bound to the agent, creates a conversation, and sends each user input as a response within that conversation [10]. Microsoft notes that "the conversation maintains history across these interactions" [10]. For the exam, understand the difference between a stateless model call and an agent conversation, and know the new terms: Conversations, Items and Responses have replaced Threads, Messages and Runs [5].

Skill group 2: text and speech solutions in Foundry

The AI-901 text and speech skill group has three skills: text analysis, spoken prompts to a multimodal model, and an application that uses Azure Speech in Foundry Tools [1].

Text analysis

AI-901 lets you approach text analysis through a general-purpose model or through Azure Language in Foundry Tools. Azure Language "provides Natural Language Processing (NLP) features for understanding and analyzing text", including personally identifiable information (PII) detection, language detection and named entity recognition [12]. Microsoft now lists sentiment analysis, key phrase extraction and summarisation under Azure Language's "Legacy capabilities", which are still supported for existing implementations [12].

That "legacy" label does not take these techniques off the exam. In the AI concepts domain, the AI-901 study guide still names keyword extraction, entity detection, sentiment analysis and summarisation as common text analysis techniques [1]. Know what each technique returns, and know that a prompted language model can now perform many of the same tasks.

Speech and spoken prompts

Azure Speech in Foundry Tools "provides speech to text, text to speech, and other capabilities through a Microsoft Foundry resource" [11]. Speech to text converts audio into text, and text to speech converts text into synthesised speech using neural voices [11]. Build at least one small app that transcribes a short audio clip and one that speaks a generated answer back.

The spoken-prompt skill is different from the Speech service skill. "Respond to spoken prompts by using a deployed multimodal model" means sending audio input to a model that accepts it, rather than transcribing first with Azure Speech [1]. The official learning path includes a dedicated speech module alongside its text analysis module [3]. Be ready to choose between a dedicated speech service and a multimodal model for a given scenario.

Skill group 3: computer vision and image generation in Foundry

The AI-901 vision skill group has three skills: interpreting images in prompts, creating new visual outputs, and building an app with vision capabilities [1].

Image understanding in Foundry now runs mainly through vision-enabled chat models. Microsoft describes these as "large multimodal models (LMM)" that "can analyze images and provide textual responses to questions about them", and the current list includes the GPT-5, GPT-4.1 and GPT-4o series and the o-series reasoning models [13]. Practise sending an image and a question in the same prompt and reading the model's description back.

Image generation in Foundry now means models such as the GPT-image series, not DALL-E. Microsoft states that "the DALL-E image generation model dall-e-3 was retired on March 4, 2026" and directs users to "a gpt-image- series model for image generation instead" [14]. The Foundry Images playground lets you try image generation without code before you call it from an application [7][14]. The official computer vision module also introduces video generation models, which the study guide does not name explicitly [24].

Skill group 4: information extraction with Azure Content Understanding

The AI-901 information extraction skill group is built entirely on Azure Content Understanding in Foundry Tools, and it has four skills: documents and forms, images, audio and video, and a lightweight application [1].

Azure Content Understanding uses generative AI to turn unstructured content into structured output. Microsoft describes it as a Foundry Tool that can "process and ingest many types of content, including documents, images, videos, and audio, into a user-defined output format" [15]. The GA (generally available) API version is 2025-11-01 [15].

Analysers are the concept to master for this skill group. An analyser "defines how your content is processed", covering content extraction settings, the field schema and the model deployments it uses, and Content Understanding offers both prebuilt and custom analysers [15]. Fields can be filled by three methods: Extract (documents only), Classify and Generate [15]. When you enable the confidence setting in a document analyser, each extracted field value gets a confidence score between 0 and 1. Output can be Markdown for search and retrieval, or JSON matching your schema [15].

Content type What you practise Content Understanding feature to know
Documents and forms Pull fields such as names, dates and totals from an invoice or form Extract method, field schema, confidence scores
Images Describe or classify what an image contains Classify and Generate methods
Audio and video Summarise a recording or pull key moments from it Generate method, Markdown output
Lightweight app Call an analyser from code and read the JSON result Prebuilt vs custom analysers

What to ignore: out-of-date AI-901 material

The biggest AI-901 trap is old material that still looks official. The "Find documentation" table on Microsoft's own AI-901 study guide still links Language Understanding (LUIS) and Anomaly Detector under old Cognitive Services paths [1]. Neither is in the AI-901 skills outline.

  • LUIS was fully retired on 31 March 2026, and Microsoft recommends migrating to conversational language understanding [17].
  • Anomaly Detector was retired on 1 October 2026, and Microsoft recommends migrating to Microsoft Fabric [18].
  • DALL-E 3 was retired on 4 March 2026 and replaced by the GPT-image models [14].
  • Hub-based projects live in the Foundry (classic) portal, while new investment goes into Foundry projects [5].

Use the study guide's skills list, not its documentation table, as your syllabus. The skills list reflects the 15 April 2026 outline [1]. Our list of common AI-901 exam mistakes covers more of the outdated vocabulary that trips candidates up.

The official learning path also includes one module that goes beyond the exam. "Get started with AI applications and agents on Azure" has seven modules, and its final module on Microsoft Foundry IQ has no matching skill in the AI-901 study guide [1][3]. It is useful background, but give it the lowest priority.

A one-week hands-on lab plan for the Foundry domain

This seven-day plan covers every AI-901 Foundry skill with roughly one to two hours of hands-on work a day. It assumes you have finished the AI concepts domain or are studying it in parallel. For a full four-week schedule across both domains, use our AI-901 30-day study plan.

Day Focus Hands-on outcome
1 Set-up, prompts and deployment Foundry project created, a Global Standard model deployed, three system prompts tested in the playground
2 Foundry SDK chat client A Python script that sends a prompt with the Responses API and prints the answer
3 Single agent A prompt agent built and tested in the agents playground, then called from code over a two-turn conversation
4 Text and speech One text analysis app (Azure Language or a model) and one Azure Speech app that transcribes or speaks
5 Vision and image generation An image question answered by a multimodal model and one image generated in the Images playground
6 Content Understanding A document analyser that extracts three fields with confidence scores, plus one audio or video summary
7 Review and timed practice The official practice assessment, then a timed practice test and a list of weak skills to revisit

The official Microsoft Learn resources map neatly onto this plan. Modules two to six of the "Get started with AI applications and agents on Azure" path cover generative AI and agents, text analysis, speech, computer vision and information extraction in turn [3]. Microsoft's instructor-led course AI-901T00-A, "Introduction to AI in Azure", runs for one day and teaches a mix of AI concepts and technology skills [23].

How to check you are ready for the Foundry domain

Microsoft now offers a free AI-901 practice assessment on AI Skills Navigator, and you must be signed in to launch it [2]. Use it early in the week to find gaps and again on day seven to confirm progress.

A realistic readiness test is whether you can complete each lab without following the tutorial line by line. If you can deploy a model, write the chat client, create an agent and run a Content Understanding analyser from memory, you have covered most of the 55-60% domain. If you still rely on copying code, repeat days two, three and six before you book. AI-901 is a fundamentals exam, but it is a harder fundamentals exam than AI-900 for anyone who has not written Python before; our guide to Microsoft AI certifications without coding is honest about what that means.

Timed, exam-style practice closes the gap between doing a lab and answering a scenario question. Labs teach you how Foundry works, while scenario questions test whether you can pick the right tool, model or deployment type quickly. Examinotion's AI-901 practice tests are written against the current outline, and the AI-901 study guide summarises both domains in one place.

Frequently Asked Questions

Does AI-901 require coding?

Yes, AI-901 expects basic Python. Microsoft's exam page says candidates need knowledge of Python coding syntax and programming techniques and should be familiar with REST APIs, SDKs and CLIs. You will not write long programs, but you must read and complete short Foundry SDK code, such as a chat client or a call to a single agent.

How much of AI-901 is about Microsoft Foundry?

The Microsoft Foundry domain makes up 55-60% of the AI-901 exam, making it the larger of the two domains. It covers 15 hands-on skills across four groups: generative AI apps and agents, text and speech, computer vision and image generation, and information extraction with Azure Content Understanding. The remaining 40-45% covers AI concepts.

Do I need an Azure subscription to study for AI-901?

You need an Azure subscription to practise the AI-901 Foundry labs properly. Microsoft says the Foundry platform is free to explore, but you need a subscription to build agents, and model calls are billed at normal rates. New customers can use the Azure free account credit, and many Foundry Tools offer a free F0 tier.

What is Azure Content Understanding?

Azure Content Understanding is a Foundry Tool that uses generative AI to turn documents, images, audio and video into structured output. You configure an analyser with a field schema, and it returns content as Markdown or extracted field values as JSON, with optional confidence scores for document fields. AI-901 names it in every information extraction skill.

Is there a free AI-901 practice assessment?

Yes. Microsoft offers a free AI-901 practice assessment on AI Skills Navigator, and the AI-901 exam page notes that you must be signed in to AI Skills Navigator to launch it. Use it to find weak skills early, then follow up with timed, scenario-style practice tests before you book the real exam.

Does AI-901 still cover LUIS or Anomaly Detector?

No. Neither LUIS nor Anomaly Detector appears in the AI-901 skills measured as of 15 April 2026. Both services are retired: LUIS on 31 March 2026 and Anomaly Detector on 1 October 2026. They still appear in the study guide's documentation table, which has not been updated, so skip those links.

Conclusion

The Microsoft Foundry domain is where AI-901 is won or lost. It carries 55-60% of the exam [1], it expects basic Python [2], and every skill rewards hands-on practice over reading. Deploy a model, build the chat client, create an agent, then work through text, speech, vision and Content Understanding, and ignore the retired services that still linger in older material.

When you are ready to test yourself under exam conditions, start practising for AI-901 with Examinotion, or browse Examinotion's exam preparation courses to compare every Microsoft AI certification we cover. For the full exam picture, read our AI-901 exam guide. If AI-901 is a stepping stone, our AI-103 guide covers the associate-level Foundry exam that follows it.

Sources

  1. Study guide for Exam AI-901: Microsoft Azure AI Fundamentals - Microsoft Learn, accessed 2026-11-09
  2. Exam AI-901: Microsoft Azure AI Fundamentals - Microsoft Learn, accessed 2026-11-09
  3. Get started with AI applications and agents on Azure (learning path) - Microsoft Learn, accessed 2026-11-09
  4. Get started with generative AI and agents in Azure (module) - Microsoft Learn, accessed 2026-11-09
  5. What is Microsoft Foundry? - Microsoft Learn, accessed 2026-11-09
  6. Deployment types for Microsoft Foundry Models - Microsoft Learn, accessed 2026-11-09
  7. Microsoft Foundry playgrounds - Microsoft Learn, accessed 2026-11-09
  8. Get started with code in Microsoft Foundry (quickstart) - Microsoft Learn, accessed 2026-11-09
  9. What is Foundry Agent Service? - Microsoft Learn, accessed 2026-11-09
  10. Quickstart: Create a prompt agent - Microsoft Learn, accessed 2026-11-09
  11. What is Azure Speech? - Microsoft Learn, accessed 2026-11-09
  12. What is Azure Language in Foundry Tools? - Microsoft Learn, accessed 2026-11-09
  13. Use vision-enabled chat models - Microsoft Learn, accessed 2026-11-09
  14. How to use image generation models - Microsoft Learn, accessed 2026-11-09
  15. What is Azure Content Understanding in Foundry Tools? - Microsoft Learn, accessed 2026-11-09
  16. What are Foundry Tools? - Microsoft Learn, accessed 2026-11-09
  17. What is Language Understanding (LUIS)? (retired) - Microsoft Learn, accessed 2026-11-09
  18. What is Anomaly Detector? (retired) - Microsoft Learn, accessed 2026-11-09
  19. Microsoft Certified: Azure AI Fundamentals - Microsoft Learn, accessed 2026-11-09
  20. Renew your Microsoft Certification - Microsoft Learn, accessed 2026-11-09
  21. Azure free account - Microsoft Azure, accessed 2026-11-09
  22. Microsoft Foundry pricing - Microsoft Azure, accessed 2026-11-09
  23. Course AI-901T00-A: Introduction to AI in Azure - Microsoft Learn, accessed 2026-11-09
  24. Get started with computer vision in Azure (module) - Microsoft Learn, accessed 2026-11-09

Preparing for a Microsoft AI Certification?

Try 5 free practice questions with detailed explanations, no credit card required.

Lifetime access280+ questions per exam7-day money-back guarantee
Start practising today

Ready to Pass Your Exam?

Don't leave your certification to chance. Prepare with exam-style practice questions, case studies, and detailed explanations for every answer.

No credit card required • Instant access