Domain 1: Fundamentals of AI and responsible AI
- AI workload types include machine learning, computer vision, natural language processing, document intelligence and knowledge mining, generative AI, and AI agents.
- Common machine learning task types: regression predicts a numeric value, classification predicts a category, clustering groups unlabeled data, anomaly detection flags outliers, and forecasting predicts future values over time.
- Features are the input variables a model learns from; labels are the known answers used in supervised learning. Models are trained on data, then used for inference (predictions) on new data.
- Datasets are split into training, validation, and test sets; overfitting means a model memorizes training data and fails to generalize to new data.
- Microsoft's six Responsible AI principles are fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability.
- Fairness avoids bias across groups, transparency makes decisions understandable, and accountability keeps people responsible for AI outcomes.
Domain 2: Fundamentals of generative AI and language models
- Large language models are built on the transformer architecture and process text as tokens; embeddings represent meaning as numeric vectors used for similarity and search.
- A prompt is the input and a completion is the model's response; the context window limits how much text the model can consider at once.
- Temperature controls randomness: lower values give focused, deterministic output while higher values give more creative, varied output.
- Prompt engineering techniques include zero-shot and few-shot prompting, system messages, and chain-of-thought prompting.
- Grounding and retrieval-augmented generation (RAG) supply the model with your own trusted data so answers stay relevant and current, which reduces hallucinations.
- Responsible generative AI uses Azure AI Content Safety to filter harmful content, Prompt Shields to defend against prompt injection and jailbreaks, and evaluation on groundedness, relevance, and coherence.
Domain 3: Build and manage AI solutions with Microsoft Foundry
- Microsoft Foundry (formerly Azure AI Foundry) is the unified platform to build, deploy, and manage AI solutions, organized into projects, hubs, and connected resources.
- The model catalog lets you browse and compare foundation models from Microsoft, OpenAI, and other providers.
- Deployment options include serverless API (pay per token, no infrastructure to manage) and managed compute (dedicated capacity you provision).
- Deployed models expose an endpoint URL and key that applications use to send prompts and receive completions.
- Foundry supports fine-tuning, model evaluation and benchmarks, prompt flow, and grounding with your own data and indexes.
- Monitoring, quota, and cost management help you operate AI solutions responsibly and within budget.
Domain 4: AI agents and agentic solutions
- An AI agent uses tools (function calling), follows orchestration steps, and acts with some autonomy to complete a goal, grounded in your data.
- Use an agent instead of a single prompt when a task needs multiple steps, external actions, or decision-making.
- The Azure AI Agent Service in Microsoft Foundry builds agents with threads, tools, and knowledge sources.
- Tool or function calling lets an agent call external functions and APIs to take real actions or fetch live data.
- Multi-agent solutions coordinate several specialized agents to solve complex tasks.
- Responsible agent design adds guardrails, human-in-the-loop checkpoints, and safety controls.
Domain 5: Computer vision, natural language, speech, and document intelligence on Azure
- Azure AI Vision provides image analysis (captions, tags, objects), OCR through the Read capability, and Face detection; Custom Vision trains your own image classification or object detection model.
- Azure AI Language provides entity recognition, sentiment analysis, key phrase extraction, PII detection, summarization, and language detection.
- Azure AI Translator provides text and document translation across many languages.
- Azure AI Speech provides speech-to-text, text-to-speech, and speech translation.
- Azure AI Document Intelligence extracts fields and tables from forms and documents using prebuilt and custom models.
- Azure AI Search indexes your content to power retrieval-augmented generation (RAG) over your own data.
AI-901 exam tips
- Focus on what each Azure AI service does and which workload fits a given scenario - that is most of the exam.
- Memorize the six Responsible AI principles by name and be able to match each to a scenario.
- Know the core generative AI vocabulary: tokens, embeddings, prompts, context window, temperature, grounding, RAG, and prompt injection.
- Understand what Microsoft Foundry is for and the difference between serverless API and managed compute deployments.
- No coding is required; the exam is conceptual.
Study guide FAQ
Is AI-900 still available?
No. Microsoft retired AI-900 on 30 June 2026 and replaced it with AI-901. Register for AI-901; both grant the same Azure AI Fundamentals credential.
Do I need coding experience for AI-901?
No. AI-901 is a beginner, conceptual exam. You are tested on what services do and when to use them, not on writing code.
What changed from AI-900 to AI-901?
The focus moved toward generative AI, AI agents, and Microsoft Foundry, and away from the older classical machine learning framing.