Certification transition

AI-102 Is Retired: What Changed in AI-103

AI-103 is not an AI-102 question bank with a new code. It moves the centre of gravity toward Microsoft Foundry, generative applications, RAG, and agents.

Reviewed August 30, 2026 · Skills measured as of April 16, 2026

The short answer

Microsoft retired AI-102 on June 30, 2026. AI-103 is the current associate certification for developers building Azure AI applications and agents. Your knowledge of Azure AI services still matters, but a study plan that ignores Foundry projects, RAG, agents, evaluation, and operational controls is incomplete.

Check the dates in the official retirement list and use the official AI-103 study guide as the source of truth.

What changed in practical terms

Preparation areaWhat to do for AI-103
Azure AI platformUnderstand Foundry resources, projects, connections, deployments, identity, networking, quotas, and monitoring as one operating environment.
Generative AIMove beyond prompt basics into model selection, RAG design, grounding, structured outputs, evaluation, safety, and observability.
AgentsPractice tools, instructions, memory, orchestration, multi-agent decisions, identity, tracing, and responsible operation.
Applied AI servicesKeep vision, language, speech, and information extraction, but learn to choose and integrate them inside modern solutions.
Question practicePrioritise scenario reasoning and current documentation. Treat material that only renames AI-102 as suspect.

What still carries over

  • Choosing the right Azure AI capability for a requirement.
  • Authentication, managed identity, role-based access, private networking, and responsible AI.
  • Computer vision, language, speech, document processing, and search fundamentals.
  • Reading Python, REST, JSON, and configuration in scenario questions.
  • Explaining trade-offs instead of memorising product names.

Do not throw away sound AI-102 knowledge. Reframe it inside the AI-103 architecture and verify every renamed service, portal workflow, SDK, role, and API against current Microsoft documentation.

Where to spend the extra study time

The largest AI-103 domain is Implement generative AI and agentic solutions at 30–35%. Plan and manage an Azure AI solution contributes another 25–30%. Together, those two areas can represent more than half of the measured skills.

  1. Build one Foundry project and be able to explain every connection and identity.
  2. Implement a small RAG flow and justify chunking, retrieval, grounding, and evaluation choices.
  3. Build an agent with at least one tool, then trace and evaluate its behaviour.
  4. Review vision, text analysis, and information extraction as solution-selection scenarios.
  5. Run timed practice and use wrong answers to build a targeted revision list.

Turn the transition into a baseline

Use ten original scenarios to see whether your current AI-102 knowledge transfers to the new blueprint.

Start the free AI-103 diagnostic

Next step

Continue with the 30-day AI-103 study plan, or review the complete AI-103 practice exam overview.