Microsoft AI-901: Complete Course Guide, All 12 Modules Explained (2026)

Microsoft has launched Course AI-901T00-A: Introduction to AI in Azure on Microsoft Learn, the foundational course that replaced AI-900 when it retired on June 30, 2026. If you are preparing for the AI-901 exam, this is the only complete guide you need. If you want to understand why hands-on practice matters more than memorization, we break that down separately.

Unlike other posts that describe only part of the course, this guide covers all 12 modules across both learning paths exactly as published on Microsoft Learn, so you know precisely what to study and what to expect.

What Is the AI-901 Course?

According to Microsoft Learn, this course "introduces fundamental concepts related to artificial intelligence (AI), and the services in Microsoft Azure that can be used to create AI solutions. It teaches a mix of AI concepts and technology skills that are considered foundational to a successful career implementing AI solutions on Microsoft Azure."

The audience profile states the course is "for aspiring technology professionals at the beginning of their career in AI solution development. Some knowledge of Python coding syntax and programming techniques is useful. Additionally, knowledge of core cloud concepts, including cloud storage, cloud compute, and cloud-based authentication and authorization is recommended."

The Two Learning Paths: 12 Modules Total

This is the key fact most guides get wrong. AI-901 consists of two separate learning paths, each containing 6 modules, for a total of 12 modules.

Description (from Microsoft Learn): "Learn core AI concepts and terminology, starting with a basic introduction and then diving into common AI workloads like: Generative AI and agents, Computer vision, Speech, Natural language processing and text analysis, Information extraction."

Prerequisites: A basic understanding of computing concepts and math. Platform: Microsoft Foundry · Role: AI Engineer · Developer

Duration: 40 minutes · Units: 10

Description: "Curious about artificial intelligence? Want to understand what the buzz is about? This module introduces you to the world of AI."

Learning objective: In this module, you learn about the kinds of solutions AI can make possible and considerations for responsible AI practices.

This opening module provides a high-level overview of the core capabilities of AI. It covers six core topic areas:

• Generative AI and agents. Generative AI enables applications to generate new content including natural language, images, video, and code. Covers LLMs vs SLMs, AI agents, content creation, translation, and summarization.
• Text and natural language. NLP foundation for generative AI. Covers text classification, sentiment analysis, key-term extraction, entity detection, and summarization.
• Speech. Speech recognition (speech-to-text) and speech synthesis (text-to-speech). Common scenarios include AI agents understanding spoken input, automated transcription, and speech translation.
• Computer vision. Image classification, object detection, semantic segmentation, and multi-modal models combining visual features with text descriptions.
• Information extraction. AI automates extraction from unstructured sources such as scanned documents, forms, and images. The basis for most solutions is optical character recognition (OCR).
• Responsible AI. All six Microsoft responsible AI principles: Fairness, Reliability and Safety, Privacy and Security, Inclusiveness, Transparency, and Accountability, with real-world examples for each.

Duration: 37 minutes · Units: 7

Description: "Generative AI powers applications that can create content, answer questions, and assist with tasks. In this module, you'll explore the fundamentals of generative AI, including large language models (LLMs), prompts, and AI agents."

Learning objectives: Describe core concepts of generative AI; explain how LLMs work; consider how to create effective prompts; describe core concepts of agents and agentic AI solutions.

• Large language models (LLMs). LLMs encapsulate linguistic and semantic relationships between words. Covers tokenization, transformer architecture with encoder and decoder blocks, attention mechanism, and embeddings, vectors containing contextual information about how tokens relate.
• Prompts. Two types: system prompts (setting behaviour, tone, constraints) and user prompts. Also covers conversation history, retrieval augmented generation (RAG), and tips for better prompts.
• AI agents. Software applications built on generative AI that can reason over natural language, automate tasks using tools, and respond to contextual conditions. Three key elements: an LLM (the brain), instructions (system prompt), and tools (knowledge + action tools). Also covers multi-agent systems.

Duration: 30 minutes · Units: 7

Description: "Natural language processing (NLP) supports applications that can analyze text to infer semantic meaning."

Learning objective: Explore concepts and techniques for text analysis.

• Tokenization. Breaking text into tokens. Pre-processing: text normalization, stop word removal, N-gram extraction, stemming, lemmatization, and POS tagging.
• Statistical text analysis. Frequency analysis, TF-IDF for relevance scoring, bag-of-words for Naive Bayes classifiers, and TextRank for extractive summarization.
• Semantic language models. Vector-based approaches (Word2Vec, GloVe). Embeddings, cosine similarity, vector arithmetic, and analogical reasoning.

Duration: 28 minutes · Units: 7

Description: "Imagine AI apps and agents that you can talk to. Explore the concepts behind AI speech, including speech recognition and synthesis."

Learning objectives: Identify speech scenarios; describe how speech recognition works; describe how speech synthesis works.

• Speech recognition, six-stage pipeline: Audio capture at 16 kHz, pre-processing using MFCCs, acoustic modeling using transformers, language modeling, decoding using beam search, and post-processing including capitalization and confidence scoring.
• Speech synthesis, four-stage pipeline: Text normalization, linguistic analysis with G2P conversion, prosody generation for pitch and duration, and audio synthesis using neural vocoders (WaveNet, WaveGlow, HiFi-GAN).

Duration: 34 minutes · Units: 9

Description: "Introduction to computer vision concepts."

Learning objectives: Identify CV task types; describe filters; describe CNNs; describe ViTs; describe generative image creation.

• Computer vision tasks. Image classification, object detection, semantic segmentation, and contextual image analysis using multimodal models.
• Images and image processing. Images as arrays of pixel values. RGB channels. Filters defined by kernels convolved across images, including the Laplace filter for edge detection.
• CNNs. Extract numeric feature maps using filters, feeding into deep learning models. Training adjusts filter weights using a softmax function for class probabilities.
• Vision transformers and multimodal models. ViT models extract patches and use attention for contextual relationships. Multimodal models combine language and vision encoders with cross-model attention.
• Image generation. Diffusion models using prompts to iteratively remove noise until producing coherent images. Similar processes apply to video generation.

Duration: 28 minutes · Units: 7

Description: "AI information extraction enables you to extract data values from documents, images, and other unstructured data sources."

Learning objectives: Understand information extraction concepts; describe how OCR extracts text; explain how form extraction maps text to data fields.

• Overview. Information extraction combines computer vision for text detection and ML or generative AI for mapping text to data fields. Scenarios: financial documents, legal and compliance, healthcare, supply chain.
• OCR, five-stage pipeline: Image acquisition, preprocessing (noise reduction, skew correction using CNNs and GANs), text region detection using LayoutLM and Mask R-CNN, character recognition using CNNs, ResNet, and BERT variants, and output generation with coordinate mapping and quality validation.
• Field extraction, five-stage pipeline: OCR output ingestion, field detection (template-based, ML, or generative AI prompt-based), field mapping using NER and proximity analysis, data normalization, and integration with business systems.

Description (from Microsoft Learn): "Learn how to get started with AI workloads and solutions in Microsoft Foundry. Start with a basic introduction and then dive into common AI workloads like: Generative AI and agents, Text analysis, Speech, Computer vision, Information extraction."

Prerequisites: A basic understanding of computing concepts and Python. Platform: Microsoft Foundry · Role: AI Engineer · Developer

This second learning path is where theory becomes practice. Each module pairs directly with its counterpart in Learning Path 1, this time showing you how to implement the same capabilities inside Microsoft Foundry.

Units: 8

Description: "Learn how Microsoft enables you to build AI with the latest technology, securely, and at scale."

• Understanding Azure. Four core service categories: Compute, Storage, Networking, and App Services. Azure organizes access through tenants, subscriptions, resource groups, and individual resources.
• Developing AI apps on Azure. Security and networking (Azure Entra ID, Key Vault, RBAC), hosting and scaling (AKS, App Service), data storage (SQL DB, Cosmos DB, PostgreSQL), and AI capabilities through Microsoft Foundry.
• Microsoft Foundry for AI. A unified enterprise-grade PaaS for building, deploying, and managing AI applications and agents. Four core capabilities: Models (GPT-5, Claude, Mistral, Llama, DeepSeek, Grok), Agents, Tools, and Knowledge via Foundry IQ.
• Foundry resources and projects. A Foundry resource provides model hosting, agent service, deployment governance, monitoring, and security boundaries. A Foundry project is a workspace for building agents, evaluations, flows, and vector indexes.

Units: 7

Description: "Learn how to use generative AI models and agents with Microsoft Foundry."

Learning objectives: Describe how an agent encapsulates model behaviour; identify the right model using the Foundry model catalog; deploy a model and replicate Playground settings in code; create and test an agent in the Foundry portal; call an agent from code using the Foundry Project API.

• Generative AI models, the model catalog. Foundry's model catalog as a central hub. Model families: GPT-5.x (multi-step reasoning), Claude Opus 4.5 (complex agents), Mistral Large 3 (multilingual, mid-latency), GPT-4.1 (speed and efficiency).
• Model evaluation. Leaderboards, side-by-side comparisons, NLP quality metrics (accuracy, precision, recall, F1), and AI-assisted quality metrics (groundedness, relevance, coherence, fluency).
• Deploying a model. Deployment parameters: type (standard, global batch, regional provisioned throughput), model version, and TPM rate limits.
• Creating an agent. Creating and testing agents in the Foundry portal, Playground for prompt validation, calling agents from code via the Foundry Project API.

Units: 7

Description: "Explore Azure Language's text analysis features, which include sentiment analysis, key phrase extraction, named entity recognition, and language detection."

• Azure Language. Capabilities: sentiment analysis (positive, neutral, negative), key phrase extraction, named entity recognition (people, places, organizations, dates), and language detection. Provisioning in Foundry.
• Azure Language SDK. Authenticating requests, sending text to the service, processing structured JSON responses.
• Azure Language MCP. Model Context Protocol integration connecting Azure Language capabilities directly to AI agents and development tools.

Units: 7

Description: "Learn how to recognize and synthesize speech using Azure Speech in Foundry Tools."

Learning objectives: Use Azure Speech's speech-to-text API; use text-to-speech API; use Voice Live capabilities.

• Speech recognition. Using the speech-to-text API: provisioning in Foundry Tools, authenticating, streaming audio input, receiving transcribed text with word-level timestamps and confidence scores.
• Speech synthesis. Converting text to natural-sounding audio: selecting neural voices, adjusting rate and pitch, generating audio in different output formats.
• Creating a speech-capable agent. Azure Speech Voice Live for real-time spoken conversations. Configuring an agent that connects speech recognition and synthesis in a unified conversational pipeline.

Units: 7

Description: "This module introduces computer vision capabilities in Foundry, focusing on how developers can analyze images and generate visual content using multimodal, image-generation, and video-generation models."

• Multimodal models for image analysis. Submitting images alongside text prompts to get descriptions, extract information, read text in images, and identify objects or scenes.
• Image generation models. Using diffusion-based models in Foundry to generate images from natural language prompts. Controlling output parameters and integrating via the Foundry API.
• Video generation models. Generating video clips from text prompts. Models account for physical behaviour and temporal progression.

Units: 6

Description: "AI gives you the power to unlock insights from your data. Learn how to use Azure Content Understanding in Foundry Tools to extract information from content."

Learning objectives: Identify Foundry Tools for information extraction; describe Azure Vision OCR capabilities; describe Azure Content Understanding; describe Azure Document Intelligence; describe Azure AI Search capabilities.

• Extract information from documents. Azure Document Intelligence for structured data extraction. Prebuilt models (invoices, receipts, ID documents), custom models, Azure Vision OCR, Azure Content Understanding for multimodal extraction, and Azure AI Search for knowledge mining.
• Extract information from audio and video. Using Azure Content Understanding to extract information from audio recordings and video files including transcription.

The biggest difference is Microsoft Foundry. AI-901 is built entirely around Foundry: the model catalog, Foundry portal, Foundry projects, Foundry Tools, Foundry IQ, and Voice Live agents are all explicitly covered. AI-900 had no Foundry content at all.

How to Access the AI-901 Course

Both learning paths are free and available now on Microsoft Learn:

• Learning Path 1: learn.microsoft.com/en-us/training/paths/ai-concepts/
• Learning Path 2: learn.microsoft.com/en-us/training/paths/get-started-ai-apps-agents/
• Full course page: learn.microsoft.com/en-us/training/courses/ai-901t00

While the conceptual modules are free to read, Microsoft's hands-on exercises cover only a fraction of what the AI-901 exam actually tests.

The official Learning Path gives you a starting point, but it was not designed to prepare you for every scenario in the exam. There are no guided labs for responsible AI evaluation, no hands-on practice for automated ML experiments, and no scenario-based exercises that mirror the kind of business problems the exam puts in front of you.

GetHandsOn.ai was built to fill that gap. Our AI-901 lab track contains hands-on scenarios that go beyond what Microsoft Learn covers, each one mapped directly to an exam objective, each one set inside a real enterprise context so the practice transfers directly to the actual test.

Should You Still Take AI-900?

No. AI-900 retired on June 30, 2026, so it can no longer be taken. If you were studying for it, the core AI concepts carry over to AI-901. See our full breakdown of the 2026 certification changes for details. If you already passed AI-900, Microsoft Fundamentals certifications have lifetime validity, so it does not expire.

AI-901 is the exam to take now. AI-900 teaches you to understand AI, while AI-901 goes further and covers how to build AI applications and agents using Microsoft Foundry. If your goal is a hands-on technical AI career, AI-901 aligns directly with 2026 job requirements.

Once you clear the AI-901 Fundamentals, your next step is the Associate-level AI-103 exam. Read our Complete AI-103 Study Guide to see what's ahead.

All course descriptions, learning objectives, module structures, and unit details are sourced directly from Microsoft Learn. Published March 30, 2026.