The current starting point
AI-900 retired on 30 June 2026. Microsoft’s Azure AI Fundamentals certification now points to AI-901. This is not a reason to buy another outdated AI-900 course package: compare the active objectives before preparing. The current AI-901 guide combines understanding AI concepts with implementing solutions in Microsoft Foundry, and its audience information expects introductory Python and Azure familiarity. Do not assume every fundamentals assessment is entirely non-technical.
Build concepts around a real decision
Begin with three fictional tasks: predict demand from historical data, classify a support message and generate a grounded answer from an approved knowledge base. Explain the different inputs and expected outputs before selecting a service. Compare generative output with a deterministic lookup; a fluent response is not proof that the underlying fact is correct. For each task, describe the possible harm from an incorrect result and who should review it.
Learn enough Python and Azure to follow the workflow
Practise reading a short script, identifying variables, calling a function and interpreting an error message. In an authorised Azure learning environment, identify the resource, project, model endpoint and permissions used by an exercise. You do not need to become a production engineer before learning fundamentals, but copying a notebook without knowing which service is called leaves a gap. Check resource charges and clean up test resources using the owner’s agreed process.
Use a three-part or three-week plan
A suggested sequence is concepts and responsible AI, Foundry implementation exercises, then integrated review. Spread it over three weeks if that matches your starting point; extend it when the practical work is unfamiliar. After a Microsoft Learn unit, explain the task in your own words and reproduce one small step. Do not treat completing every progress bar as evidence that you can diagnose a new scenario.
Test responsible AI in the example
Create a small test set containing an ambiguous request, missing information and an inappropriate request. Record the expected behaviour before examining the output. Check whether the answer is supported, whether private data is exposed and whether uncertainty is communicated. This is an educational quality exercise, not a claim that a handful of examples certifies a production system as safe. Production decisions need the organisation’s formal review and broader evaluation.
Readiness and the next career step
Use the current official practice assessment and the exam sandbox to assess knowledge and interface familiarity separately. When a question is wrong, connect it to a concept or task you can explain, rather than memorising the option. AI-901 can help structure an early learning path, but it does not alone establish AI engineering competence. Consider development, data or business AI routes according to your intended work, then read their current requirements before paying for the next credential.
Official references
Certification requirements and programme terms can change. Use the issuer’s current page and your account to confirm details before applying or paying.