Quick one to flag if you're on the Azure AI track. AI-102, the exam behind the Azure AI Engineer Associate certification, retired back on June 30. Its replacement, AI-103, leading to the new Microsoft Certified: Azure AI Apps and Agents Developer Associate credential, is live and bookable through Pearson Vue right now.
The name change is the tell. AI-102 was framed around "AI Engineer," a fairly broad title covering cognitive services, vision, language, and search. AI-103 narrows and modernizes that scope around Microsoft Foundry and agentic AI specifically. The official skills list covers planning and managing an Azure AI solution, implementing generative AI and agentic solutions, computer vision, text analysis, and information extraction. Notice that "agentic solutions" sits second on that list, not buried at the bottom. Microsoft is explicitly testing whether you can build things that act, not just things that respond.
What's actually different from AI-102
If you've looked at AI-102 material before, a lot of the underlying skills carry over. You still need real Python experience and working familiarity with Azure AI services. What's new is the framing around Microsoft Foundry as the development hub, plus explicit coverage of tool integration, grounding, and responsible AI practices for agents specifically, things like making sure an autonomous agent doesn't take an action it shouldn't. It's a 120 minute proctored exam with an interactive component, so expect more scenario based questions than straight multiple choice.
This is worth mentioning without dwelling on it: it's the same directional shift we've flagged with other Microsoft exam updates recently, cert content moving from "do you know the service" to "can you build and govern something that uses AI autonomously." AI-103 is just the clearest single example of it so far.
Worth knowing even if you're not on Azure
Zooming out from certs for a second. AWS Interconnect for multicloud hit general availability back in April, and it's worth a mention here because of what it signals about where cloud infrastructure work is heading. It's a jointly built private connection between AWS and Google Cloud, with Microsoft Azure and Oracle Cloud both slated to join later this year. Instead of stitching multicloud networking together by hand with VPNs and manual peering, you provision it as another resource in your Terraform or Pulumi state file, with connectivity live in minutes instead of a multi week networking project.
That matters for study priorities too. If your organization runs workloads across more than one cloud, which is increasingly the default rather than the exception, "I know how to secure and provision cross cloud connectivity as code" is becoming a real, listable skill, not just an obscure networking topic. Worth a look even if it's not on any current exam blueprint yet.
What to actually do
If you're currently working toward AI-102 material, stop and switch to the AI-103 study guide, there's no reason to prep for a retired exam. If you already hold the Azure AI Engineer Associate cert, you're fine, it stays valid on its normal renewal schedule, but know your next-level path forward runs through AI-103, not a direct upgrade exam.
And if agentic AI still feels like a vague buzzword to you rather than something you could describe the guardrails for, that's the actual gap AI-103 is testing for. Spend a weekend building one small agent workflow in Microsoft Foundry, or the equivalent in Bedrock or Vertex AI if you're on a different platform. The concept transfers even if the tooling doesn't.
Keep Learning and Building.