Azure ships April SDK updates

Microsoft’s April Azure developer drop was not one big product launch. It was a stack of smaller moves that matter if you actually build on Azure every day. The headline items were a critical security fix in the Java Cosmos DB SDK, a stable 2.0 release for the.NET package behind AI Foundry projects, and general availability for the Java AI Agents library. That sounds routine. But basically it shows where Azure is pushing hardest now — safer data plumbing underneath, and cleaner agent-building tools on top. ### What actually shipped? The main release landed on April 21, 2026 in the Azure SDK monthly update. Microsoft highlighted Cosmos DB 4.79.0 for Java, Azure.AI.Projects 2.0.0 for.NET, and Azure AI Agents 2.0.0 for Java. It also listed a batch of new stable and beta packages for provisioning and management libraries, but those three are the ones with the biggest practical impact for app teams. (devblogs.microsoft.com) ### Why is the Cosmos DB fix the real news? Because this one was security, not just cleanup. Microsoft said the Java Cosmos DB library included a critical fix for a remote code execution vulnerability tied to Java deserialization. The patch replaces deserialization with JSON-based serialization in several internal components, which removes that whole attack class instead of just patching one narrow hole. That is the kind of change developers usually want to take fast, even if they ignore most monthly SDK notes. (devblogs.microsoft.com) ### What changed in AI Foundry 2.0? The.NET package Azure.AI.Projects hit stable 2.0.0 with breaking changes that look annoying at first but make the SDK more predictable. Evaluations and memory operations moved into separate namespaces. Several types were renamed — like `Insights` to `ProjectInsights` and `Schedules` to `ProjectSchedules`. Boolean properties were also normalized to `Is*` naming. In plain English, Microsoft is trying to make the AI Foundry developer surface less messy before more teams standardize on it. (devblogs.microsoft.com) ### And the AI Agents 2.0 release? That one is about Java developers getting the “stable enough for production” signal. The Java Azure AI Agents library reached general availability with version 2.0.0. Microsoft also changed enum handling, renamed `*Param` classes to `*Parameter`, fixed casing around MCP connector identifiers, and added a convenience overload for updating memories. None of that is flashy. But it is exactly the kind of API consistency work you do when you want enterprises to stop treating agents as a demo stack. (devblogs.microsoft.com) ### Where does the database program fit in? On the same day, Microsoft introduced Azure Accelerate for Databases. The pitch is simple — help companies modernize databases for AI with Microsoft delivery support, partner help, assessments, tooling, skilling, Azure credits, and database savings plans. Microsoft says customers can save up to 35% versus pay-as-you-go with the savings plan for databases, and the program is framed as a way to reduce the cost and execution risk of modernization. (devblogs.microsoft.com) ### Why bundle SDK work with modernization messaging? Because Azure’s AI story depends on boring database work getting finished. Agents need memory, retrieval, governance, and live access to operational data. If customer data is fragmented or stuck on legacy systems, the fancy agent layer does not matter much. Microsoft is now selling both halves together — developer SDKs for the application layer, and a modernization program for the data layer underneath. (azure.microsoft.com) ### Is this part of a bigger Foundry push? Yes. Microsoft’s broader Foundry updates in 2026 have leaned toward production tooling — hosted agents, observability, memory features, toolbox-style tool management, and SDK updates across languages. So the April Azure SDK release fits a larger pattern. The company is trying to make “build an agent on enterprise data” feel like normal software work, not a pile of custom glue. (azure.microsoft.com) ### Bottom line? The April SDK notes look small, but the direction is pretty clear. Microsoft is tightening security in core data libraries, stabilizing its AI developer APIs, and wrapping database modernization into the same enterprise AI pitch. That is less exciting than a new frontier model. But for Azure customers, it is probably the more important work. (devblogs.microsoft.com 1) (devblogs.microsoft.com 2)