AI on .NET & Azure: Weekly Roundup — Week of March 22, 2026

TL;DR — It was a big week in the Microsoft AI stack. GitHub Copilot gained GPT-5.4 mini, the MCP C# SDK hit v1.0, Microsoft Agent Framework is sprinting toward GA, Azure is about to light up NVIDIA Vera Rubin NVL72 racks, and Microsoft reshuffled Copilot leadership to bet everything on agentic workflows. Oh, and Windows is quietly un-AI-fying a few apps. Let’s unpack what matters for engineers shipping on .NET and Azure.

1. GPT-5.4 & GPT-5.4 mini Land in GitHub Copilot

The model carousel kept spinning this week. GPT-5.4 mini — OpenAI’s latest fast version of their agentic coding model — began rolling out in GitHub Copilot, and in early tests it appears to be OpenAI’s highest-performing mini model yet. For the average developer this means faster time-to-first-token and snappier codebase exploration without burning through a budget model’s quota. It delivers the fastest time to first token, is stronger at codebase exploration, and is especially effective with grep-style tools — though it launches with a 0.33× premium request multiplier, and pricing is tentative and subject to change. Budget accordingly.

On the enterprise side, GPT-5.3-Codex is GitHub’s first LTS model, in partnership with OpenAI — it launched February 5, 2026, and will remain available through February 4, 2027 for Copilot Business and Copilot Enterprise users. This is genuinely useful news for security teams that need a stable, audited model reference for compliance reviews. On May 17, 2026, GPT-5.3-Codex becomes the base model for all Copilot Business and Copilot Enterprise organizations — so mark your calendars and test your prompts before then.

Practical steps:

Enable GPT-5.4 mini in your VS Code model picker (requires VS Code ≥ v1.104.1 or Visual Studio ≥ 17.14.19).
Enterprise/Business admins: enable the GPT-5.4 mini policy in Copilot settings; once enabled, users will see GPT-5.4 mini in the model picker in Visual Studio Code.
Pin GPT-5.3-Codex as your org’s LTS baseline now, before the May cutover.

2. MCP C# SDK Hits v1.0 — Your Agent Toolkit Just Got Stable

If you’ve been waiting for a stable foundation before wiring Model Context Protocol into your ASP.NET services, the wait is over. The MCP C# SDK has reached its v1.0 milestone, bringing full support for the 2025-11-25 version of the MCP Specification, with a rich set of new capabilities — from improved authorization flows and richer metadata, to powerful new patterns for tool calling, elicitation, and long-running request handling. This pairs neatly with the AI Toolkit for VS Code March update. Version 0.32.0 is packed with new capabilities designed to help you ship production-ready AI agents, bringing a unified tree view experience, Agent Builder enhancements, and streamlined GitHub Copilot integration for agent development. Worth noting for tooling housekeeping: the Foundry sidebar will retire on June 1st, 2026, and all of its functionalities have been moved into the AI Toolkit sidebar. Update your team’s VS Code extension list accordingly — no one wants a surprise retirement notice on a Monday morning.

3. Microsoft Agent Framework RC: Agentic .NET, Finally Stable-ish

Microsoft has announced that the Microsoft Agent Framework has reached Release Candidate status for both .NET and Python. It succeeds earlier efforts such as Semantic Kernel and AutoGen, consolidating agent creation, orchestration primitives, and multi-provider support under a single SDK. This framework is the recommended approach for .NET apps that need to build agentic AI systems with advanced orchestration, multi-agent collaboration, and enterprise-grade security and observability — a production-ready, open-source framework that brings together the best capabilities of Semantic Kernel and Microsoft Research’s AutoGen. Agent Framework provides multi-agent orchestration with support for sequential, concurrent, group chat, handoff, and magentic patterns, and is cloud-agnostic (containers, on-premises, or multi-cloud) and provider-agnostic using plugin and connector models. A new Agent Skills system landed alongside the RC. You can now equip your Microsoft Agent Framework agents with portable, reusable skill packages that provide domain expertise on demand — without changing a single line of your agent’s core instructions, with built-in skills providers for both .NET and Python. Quick-start snippet — bootstrap a minimal agent in a new console app:

dotnet new console -o HelloWorldAgents
cd HelloWorldAgents
dotnet add package Microsoft.Agents.AI --prerelease
dotnet add package OpenAI
dotnet add package Microsoft.Extensions.AI.OpenAI --prerelease
dotnet add package Microsoft.Extensions.AI

And in an ASP.NET Minimal API, wire up your IChatClient and register an agent:

builder.AddOpenAIClient("chat")
       .AddChatClient(Environment.GetEnvironmentVariable("MODEL_NAME") ?? "gpt-4o-mini");

dotnet add package Microsoft.Agents.AI.Hosting --prerelease

Then register via builder.AddAIAgent(...). See the devblogs.microsoft.com Agent Framework blog for the full walkthrough. With this RC release, the framework’s APIs and workflows are locked down, allowing teams to start production evaluation and implementation with greater confidence. That’s as close to “ship it” as a prerelease label allows.

4. .NET 11 Preview 2 & March Servicing Drops

.NET 11 Preview 2 is now available, covering updates to the runtime, SDK, libraries, ASP.NET Core, Blazor, .NET MAUI, F#, and more. The headline for observability fans: ASP.NET Core gains native OpenTelemetry tracing support — the framework now adds OpenTelemetry semantic attributes directly to HTTP server activity, meaning developers no longer need the separate instrumentation library to collect standard tracing data. One fewer NuGet package to babysit. 🎉

On the maintenance front, .NET and .NET Framework received their March 2026 servicing update as of March 10, 2026, containing security and non-security fixes. Patch early, patch often.

5. Azure Gets Its First NVIDIA Vera Rubin NVL72 Rack

If you care about inference cost curves (and you should), this is the big infrastructure story of the month. Microsoft announced expanded Microsoft Foundry capabilities to build, deploy, and operate production-ready AI agents on NVIDIA accelerators and open NVIDIA Nemotron models, along with new Azure AI infrastructure optimized for inference-heavy, reasoning-based workloads — including being the first hyperscale cloud to power on next-generation NVIDIA Vera Rubin NVL72 systems. In less than a year, Microsoft has deployed hundreds of thousands of liquid-cooled Grace Blackwell GPUs across its global datacenter footprint, and the Vera Rubin NVL72 will be rolled out into modern, liquid-cooled Azure datacenters over the next few months. NVIDIA Vera Rubin Superchips will deliver 50 PF NVFP4 inference performance per chip and 3.6 EF NVFP4 per rack — a five times jump over NVIDIA GB200 NVL72 rack. For developers running large reasoning models via Azure AI Foundry, this translates to meaningfully lower per-token latency as capacity scales. Don’t rewrite your SLA contracts just yet, but it’s a reasonable tailwind to plan around for H2 2026.

6. Copilot Org Restructure & the “Less-is-More” Windows Pivot

Two organizational signals worth tracking:

Leadership unification. Microsoft says a new era of productivity is emerging as AI experiences rapidly evolve from answering questions and suggesting code to executing multi-step tasks with clear user control points — citing Copilot Tasks, Copilot Cowork, agentic capabilities in Office, and Agent 365. To that end, Microsoft is bringing the Copilot system across commercial and consumer together as one unified effort spanning four pillars: Copilot experience, Copilot platform, Microsoft 365 apps, and AI models. For developers building on Copilot extensibility APIs, this consolidation should mean fewer divergent SDK paths — eventually.

Windows AI pullback. Microsoft announced a series of changes focused on improving Windows 11 quality, which notably includes dialing back the number of entry points to its AI assistant, Copilot — reducing Copilot AI integrations in Photos, Widgets, Notepad, and Snipping Tool. Under the heading of “integrating AI where it’s most meaningful,” Microsoft is becoming more intentional about “how and where Copilot integrates across Windows,” aiming to focus on AI experiences that are “genuinely useful.” The lesson for product engineers: feature surface ≠ feature value.

7. Azure Language Studio Retirement Countdown

Mark your migration tickets: Azure Language Studio is scheduled for retirement on March 20, 2027. All Azure AI Language capabilities are now accessible in Microsoft Foundry (classic), delivering a complete, unified development experience — including project management, model training, and deployment workflows. Also of note for .NET devs: a new Azure Language .NET SDK preview with support for the 2025-11-15-preview API is now available. Update your NuGet references and start testing against the Foundry endpoint sooner rather than later.

What’s Coming Next

Timeline	Thing to Watch
Late March 2026	Copilot Cowork broader rollout via Microsoft Frontier program
May 1, 2026	Agent 365 GA at $15/user/month — your enterprise governance story for agents
May 17, 2026	GPT-5.3-Codex becomes default base model for Copilot Business/Enterprise
June 1, 2026	Foundry VS Code sidebar retires — migrate to AI Toolkit
Nov 2026 (est.)	.NET 11 GA
March 20, 2027	Azure Language Studio retirement