Microsoft Agent Framework 1.0 Unifies SDKs, Not the Azure Stack

What shipped

Microsoft released version 1.0 of its open-source Agent Framework on April 3, 2026. The framework merges two toolkits the company had maintained side by side, Semantic Kernel and AutoGen, into a single SDK for building AI agents. 21 Microsoft calls it the production-ready milestone for a project first introduced in October 2025. It ships for both .NET and Python with stable APIs and long-term support. 3

The pitch is aimed at developers who want more than a single chat assistant. Microsoft frames the framework as a base for orchestrated, multi-agent systems that call tools, work across multiple models, and run longer workflows. 3 Alongside the stable core, the company highlighted preview features, including a developer UI called DevUI, hosted agent integration, and deeper tooling and observability support. 3

Why consolidation was overdue

The unification fixes a split that had dogged Microsoft's agent strategy. Semantic Kernel served enterprise developers who wanted stable connectors and telemetry. AutoGen drew researchers experimenting with multi-agent orchestration. The two had incompatible APIs, so teams had to pick one and accept its tradeoffs. 2

Visual Studio Magazine describes the new framework as building on both projects rather than discarding them. It takes its foundations from Semantic Kernel and its orchestration concepts from AutoGen. 3 For developers who hedged by learning both, that continuity matters, because the ideas carry forward even where the code does not.

The lineage problem

The history behind the release is more tangled than one product replacing two others. The original AutoGen has been in maintenance mode since October 2025. It gets bug and security fixes but no new features, and its last major Python release was v0.7.5. 1 AutoGen's GitHub repository now points new users to Agent Framework. 2 Semantic Kernel has also moved to maintenance mode as of the 1.0 release. Existing plugins, planners, and memory components keep working, but new agent capabilities will land in Agent Framework instead. 1

A third branch complicates things further. AutoGen's original creators left Microsoft and forked the project as AG2, which is developed under the ag2ai organization and is heading toward its own 1.0. 1 Despite the shared ancestry, AG2 is a separate codebase from both legacy AutoGen and Microsoft's new framework, and teams should evaluate it on its own terms. 1

In practice, "we use AutoGen" could now mean any of three different things. Teams running legacy AutoGen in production are effectively patching a frozen dependency and should plan a migration. 1

The critique: fixing the SDK, not the stack

The sharpest pushback comes from Forbes. It credits the 1.0 release with solving a real problem at the framework layer but argues the bigger issue remains. In Forbes' view, Microsoft's agent offering is still spread across too many surfaces covering build, deployment, governance, and distribution. 2 By contrast, it says Google Cloud and AWS each offer a clearer default path from framework to managed runtime. 2

The distinction is useful. An SDK determines how developers write agent logic. A platform determines where that logic runs, how it is secured and monitored, and how it reaches users. Merging Semantic Kernel and AutoGen removes a decision at the first step but leaves the later steps as complicated as before.

The preview features in the 1.0 announcement, especially hosted agent integration and observability tooling, look like early efforts to connect the framework to the rest of Microsoft's stack. 3 Because they are still in preview, they have not yet reached the point where a developer can follow one obvious path from local code to governed production deployment.

Reading the release

The sources agree on the facts: one framework, two predecessors in maintenance mode, and stable APIs on two languages. 123 They differ in emphasis. Visual Studio Magazine treats the release as a milestone in maturity. 3 Forbes sees it as a necessary but partial fix. 2 The drel.ai review focuses on what teams already running these frameworks should do now. 1

The Forbes reading seems closest to the mark, though it should not be overstated. Agent Framework 1.0 is a real improvement, and committing to long-term support on both .NET and Python is what enterprise buyers needed before betting on it. Still, the competition among cloud providers is about the full journey from prototype to production, not about which SDK is most elegant. Microsoft has cleaned up the first stage of that journey.

The next test is whether the hosted and observability previews turn into an integrated, default runtime story. Until then, developers on Azure get a cleaner way to write agents but still face several choices about where to run them.