Choose a Microsoft AI credential and practice building, securing, or operating a workload in Microsoft Foundry.

Microsoft Certified: Azure AI Fundamentals (AI-901)
Start here if you are learning AI workload concepts and basic implementation in Microsoft Foundry. Practice deploying a model and calling it from a small Python client.
Microsoft Certified: Azure AI Apps and Agents Developer Associate (AI-103)
Choose this route if you develop Azure AI apps and agents. Practice grounding responses, integrating a scoped tool, and evaluating failure cases.
Microsoft Certified: Machine Learning Operations Engineer Associate (AI-300)
Choose this route if you operate ML models and generative AI workloads. Practice reproducible releases, evaluation gates, monitoring, and rollback.
Microsoft Certified: Cloud and AI Security Engineer Associate (SC-500)
Choose this route if you secure AI and cloud infrastructure. Practice limiting identities and network access, protecting data, and monitoring workload activity.
Choose a role before you choose an exam
AI-901 gives beginners a small application to build and explain. AI-103 goes deeper into developing Foundry apps and agents. AI-300 fits someone responsible for release pipelines and ongoing model performance. SC-500 focuses on protecting the surrounding cloud and AI infrastructure. Use the issuer’s current study guide as a checklist for the role you want.
Build one assistant with four perspectives
Start with a Python client and a Foundry model deployment. Add approved documents with Azure AI Search and test a scoped identity. As the developer, check grounded answers and safe tool use. As the operator, make a release gate and rehearse rollback. As the security engineer, test denied access and inspect the logs for the attempt.
Hand off a release someone else can operate
Include a diagram of identities and data flows, an evaluation report, and a runbook for access failures, cost spikes, and degraded answers. Record where diagnostic data goes and who can read it. Use a separate sandbox and tear down billable resources after testing.
Build something you can show
Build a Microsoft Foundry assistant that searches a small approved document collection. Use Microsoft Entra identity and test denied access with a second account. Add an evaluation set that covers retrieval failures and prompt injection, and document the conditions under which you would roll back a new model or prompt.