Microsoft AI-901 Azure AI Fundamentals Study Guide

---

1 What Is AI-901 and Who It Is For Beginner

AI-901 is the Microsoft Azure AI Fundamentals certification exam. Passing it earns the Microsoft Certified: Azure AI Fundamentals credential. According to Microsoft Learn, the exam is designed for candidates at the beginning of their career in AI solution development who have conceptual knowledge of AI solutions in Azure and the foundational technical skills to work with them.

Unlike its predecessor AI-900, which required no technical background, AI-901 explicitly requires knowledge of Python coding syntax and programming techniques, and familiarity with Azure resources. This is a significant audience shift: AI-901 targets practitioners who want to build and implement AI solutions, not only those who need to understand AI concepts at a business level.

The certification is appropriate for data professionals, developers, IT professionals, and anyone working with Azure who wants to demonstrate foundational AI implementation skills. It serves as a stepping stone toward more advanced Microsoft certifications including Azure AI Engineer Associate and Azure Data Scientist Associate, though it is not a prerequisite for either.

2 How AI-901 Differs from AI-900 Beginner

AI-900 was the predecessor exam, retired on June 30, 2026. AI-901 replaces it with a substantially different focus.

Area	AI-900 (Retired June 30, 2026)	AI-901 (Current)
Technical prerequisites	None required	Python coding knowledge and Azure resource familiarity required
Primary platform	Azure Cognitive Services, Azure Machine Learning	Microsoft Foundry (Foundry portal and Foundry SDK)
Implementation weight	Concept-focused	55 to 60 percent of exam is implementation in Foundry
Generative AI coverage	Introduced generative AI concepts	Deeper: deploy models, build chat apps, create agents
Information extraction	Form Recognizer and Document Intelligence	Azure Content Understanding in Foundry Tools
Practice assessment	Available	Not yet available as of July 2026. Microsoft states it is typically available within 8 weeks of GA.

3 Exam At a Glance Beginner

4 Prerequisites Beginner

Microsoft Learn specifies the following for AI-901 candidates. These are stated requirements in the official study guide, not optional background knowledge.

• Python coding syntax and programming techniques. AI-901 requires hands-on implementation in Foundry using the Foundry SDK. Reading and writing basic Python is necessary for the implementation domain.
• Azure resource familiarity. Deploying models, working with the Foundry portal, and configuring Azure services requires comfort navigating the Azure portal and understanding Azure resource concepts such as subscriptions, resource groups, and service endpoints.
• Conceptual knowledge of AI solutions in Azure. Understanding what AI workloads exist, what problems they solve, and which Azure services address them.

5 Domain 1: Principles of Responsible AI Beginner

Responsible AI is the foundation of the first domain. Microsoft defines six principles that govern how AI systems should be designed, built, and operated. All six are explicitly listed in the AI-901 study guide objectives and are examinable.

Fairness

AI systems should treat all people equitably. Candidates should understand how bias can enter AI systems through training data, how fairness is defined and measured across different demographic groups, and why fairness must be considered throughout the AI development lifecycle rather than only at deployment.

Reliability and Safety

AI systems should perform reliably and safely. Candidates should understand how AI systems can fail, the importance of testing across diverse scenarios and edge cases, and the role of human oversight in safety-critical AI applications.

Privacy and Security

AI systems should be secure and respect privacy. Candidates should understand how AI systems handle personal data, the risks of data exposure through model outputs, and how to apply privacy-by-design principles when building AI solutions.

Inclusiveness

AI systems should empower everyone and engage people across a range of needs and experiences. Candidates should understand accessibility requirements and how to design AI solutions that serve users with diverse abilities, languages, and contexts.

Transparency

AI systems should be understandable. Candidates should understand why explainability matters, what it means for an AI decision to be interpretable versus opaque, and how transparency supports user trust and accountability.

Accountability

People should be accountable for AI systems. Candidates should understand governance frameworks, the role of human review in automated decisions, and how accountability is structured in organizations deploying AI solutions.

6 Domain 1: AI Model Components and Configurations Beginner

This section tests understanding of how generative AI models function and how to select and configure them appropriately for a given task.

How generative AI models work

Candidates should understand the transformer architecture at a conceptual level: how models are trained on large datasets, how they generate outputs based on probability distributions over tokens, and how concepts like temperature and context window affect generation behavior. Knowledge of foundational concepts such as tokenization, embeddings, and attention mechanisms at an awareness level is expected.

Selecting an appropriate AI model based on capabilities

Candidates should be able to identify which type of model is appropriate for a given scenario. Key distinctions include: language models for text generation and analysis, multimodal models for combined text and vision tasks, speech models for audio processing, and specialized models for computer vision tasks. Matching capability to use case is a core exam skill.

Model deployment options and configuration parameters

Candidates should understand deployment options available through Microsoft Foundry including the Foundry portal deployment workflow. Key configuration parameters include model version selection, capacity allocation, and connection endpoints. Understanding the difference between deployment options such as standard provisioned and global standard deployments is relevant.

7 Domain 1: AI Workloads Beginner

This section covers identifying the right AI workload type for a given scenario and understanding the capabilities within each workload category.

Common AI workloads

Candidates must be able to identify appropriate scenarios for each workload type. According to the official objectives these are: generative and agentic AI, text analysis, speech, computer vision, and information extraction.

Text analysis techniques

The study guide explicitly lists the following text analysis techniques as examinable:

• Keyword extraction: identifying the most relevant terms in a piece of text
• Entity detection: identifying and classifying named entities such as people, places, organizations, and dates
• Sentiment analysis: determining the emotional tone of text as positive, negative, or neutral
• Summarization: condensing longer text into key points

Speech recognition and synthesis

Candidates should understand the capabilities and use cases for both directions of speech processing. Speech recognition converts spoken audio to text. Speech synthesis converts text to spoken audio. Candidates should be able to identify scenarios where each is appropriate and understand the Azure Speech service capabilities available through Foundry Tools.

Computer vision and image generation

Candidates should identify the features and capabilities of computer vision models including object detection, image classification, and optical character recognition. For image generation models, candidates should understand what generative image models produce and the difference between image analysis and image generation tasks.

Information extraction techniques

Candidates should identify techniques to extract structured information from unstructured content across multiple modalities: text documents, images, audio, and video. This objective maps directly to the implementation work with Azure Content Understanding covered in Domain 2.

8 Domain 2: Generative AI Apps and Agents Intermediate

This is the largest section of the exam. Microsoft Foundry is the unified platform for building AI solutions on Azure. The Foundry portal is the web-based interface. The Foundry SDK provides programmatic access through Python. Both are tested.

Creating effective system and user prompts

Candidates should understand the role of the system prompt in establishing model behavior, persona, and constraints. User prompts carry the actual task or query. Candidates should be able to write prompts that produce consistent, appropriate outputs and understand techniques such as few-shot prompting and chain-of-thought prompting at a conceptual level.

Deploying a model and interacting with it in the Foundry portal

Candidates should be able to navigate the Foundry portal to deploy a language model, configure deployment settings, and use the chat playground to test the deployed model interactively. This is a hands-on objective requiring familiarity with the actual Foundry portal interface.

Creating a lightweight chat client application using the Foundry SDK

Candidates should be able to write a basic Python application that connects to a deployed model using the Foundry SDK, sends messages to the model, and handles the response. This requires Python knowledge and understanding of the SDK’s authentication and messaging patterns.

Creating and testing a single-agent solution in the Foundry portal

Candidates should understand what an AI agent is: an AI system that can take actions, use tools, and complete multi-step tasks autonomously. The exam tests the ability to configure a single agent in the Foundry portal, define its tools and instructions, and test its behavior.

Creating a lightweight client application for an agent

Building on the agent configuration objective, candidates should be able to write a basic Python application that communicates with a configured agent through the Foundry SDK.

9 Domain 2: Text and Speech Solutions Intermediate

Building a lightweight text analysis application

Candidates should be able to implement a basic application that uses text analysis capabilities through Foundry. This includes calling services that perform the text analysis techniques listed in Domain 1: keyword extraction, entity detection, sentiment analysis, and summarization.

Responding to spoken prompts using a deployed multimodal model

Multimodal models accept multiple input types. Candidates should understand how to configure a deployed multimodal model to accept audio input and generate appropriate text responses, implementing a basic voice-to-response workflow.

Building a lightweight application using Azure Speech in Foundry Tools

Azure Speech capabilities are accessible through Foundry Tools. Candidates should understand how to build a basic application that uses Azure Speech for speech-to-text or text-to-speech functionality, accessed through the Foundry platform.

10 Domain 2: Computer Vision and Image Generation Intermediate

Interpreting visual input using a deployed multimodal model

Candidates should be able to submit image inputs to a deployed multimodal model through the Foundry portal or SDK and interpret the model’s analysis of that image. This includes scenarios such as describing image content, identifying objects, or answering questions about a visual input.

Creating new visual outputs using generative models

Candidates should understand how to use generative image models deployed through Foundry to produce new images from text prompts. This includes understanding prompt construction for image generation and the types of models available for image generation tasks in Azure.

Building a lightweight application with vision capabilities

Candidates should be able to write a basic Python application that submits image data to a vision model through the Foundry SDK and processes the resulting output.

11 Domain 2: Information Extraction with Content Understanding Intermediate

Azure Content Understanding is the service for extracting structured information from unstructured content. It is accessed through Foundry Tools. The exam tests four content types across this objective.

Extracting information from documents and forms

Candidates should understand how Content Understanding processes documents and forms to extract structured data such as field values, tables, and key-value pairs. This covers scenarios such as invoice processing, form data extraction, and document parsing.

Extracting information from images

Candidates should understand how Content Understanding extracts structured information from images including text extraction through optical character recognition, object identification, and structured data extraction from image-based documents such as photographs of paper forms.

Extracting information from audio and video

Candidates should understand how Content Understanding processes audio and video content to extract information such as transcriptions, speaker identification, key moments, and structured summaries.

Building a lightweight information extraction application

Candidates should be able to write a basic Python application that uses Content Understanding through the Foundry SDK to submit content and retrieve extracted information in a structured format.

12 Study Strategy and Official Resources Beginner

Microsoft recommends training and hands-on experience before sitting the exam. The following approach is grounded in the official Microsoft Learn paths listed in the AI-901 study guide.

Recommended study order

1. Start with the official AI-901 study guide at learn.microsoft.com and review the full objectives list. Use the weightings (40 to 45 percent for Domain 1, 55 to 60 percent for Domain 2) to prioritise study time.
2. Complete the Microsoft Learn self-paced modules for AI concepts and Microsoft Foundry. These are free and aligned to the exam objectives.
3. Get hands-on in the Foundry portal. Domain 2 is implementation-focused. Reading about Foundry is insufficient preparation. Deploying a model, building a basic chat application, and configuring an agent in the actual portal are core skills tested.
4. Set up a Python environment and practice the Foundry SDK. Lightweight applications in Python are a specific exam objective. Writing and running the SDK code in a real environment before exam day builds the familiarity needed to answer implementation questions accurately.
5. Review responsible AI principles thoroughly. The six principles from Domain 1 are conceptual but frequently tested through scenario-based questions. Understanding which principle applies to a given real-world AI failure scenario is a common question pattern.
6. Use the exam sandbox at aka.ms/examdemo to familiarise with the exam interface before the actual exam day.

Official learning paths from Microsoft Learn

• AI concepts learning path
• Get started with AI apps and agents
• AI-901 exam details page

References

• Microsoft Learn: Official Study Guide for Exam AI-901 (updated April 15, 2026)
• Microsoft Learn: Exam AI-901 Details and Registration
• Microsoft Learn: Microsoft Certified: Azure AI Fundamentals
• Microsoft Learn: AI Concepts Learning Path
• Microsoft Learn: Get Started with AI Apps and Agents
• Microsoft Learn: Exam Scoring and Score Reports
• Microsoft Learn: Certification Renewal
• Microsoft: Exam Sandbox (explore exam environment before test day)