TL;DR

Microsoft Foundry quietly shipped a meaningful SDK cleanup and capability bump in the last 48 hours. The azure‑ai‑projects 2.0.0 release (with a matching .NET 2.0.0) collapses agent APIs into a single surface, removes extra dependencies, and pairs nicely with new high‑performance open‑model inference options (Fireworks AI) now visible in Foundry. If you’re shipping AI features on .NET + Azure, this is a “stop what you’re doing and skim the changelog” moment.

The one story that matters this week

On April 9–10, 2026, Microsoft published the “What’s new in Microsoft Foundry | March 2026” update, and despite the month label, the impact is very current for engineers integrating Foundry today (devblogs.microsoft.com). The headline change isn’t a flashy model—it’s API consolidation that directly affects how you build, test, and ship agentic apps.

Why this is a big deal

Foundry’s SDK surface had started to sprawl: separate packages for projects, agents, identity, and OpenAI wiring. Version 2.0.0 tightens this up and makes agents a first‑class citizen—especially relevant if you’re building multi‑agent workflows or embedding Foundry into existing .NET services.

What actually changed (and why you should care)

1) One client to rule them all

Agents are now first‑class operations on AIProjectClient. The standalone azure-ai-agents dependency is gone.

Before (simplified):

using Azure.AI.Agents;
using Azure.Identity;

var agentClient = new AgentClient(endpoint, new DefaultAzureCredential());

After (.NET SDK 2.0.0):

using Azure.AI.Projects;

var project = new AIProjectClient(endpoint, credential);
var agent = await project.Agents.CreateAsync(...);

Fewer abstractions, fewer packages, fewer version conflicts. Your CI pipeline will thank you.

(devblogs.microsoft.com)

2) Dependency diet: install less, break less

As of azure‑ai‑projects 2.0.0, the Python package bundles openai and azure-identity directly; the .NET release mirrors the spirit by eliminating extra agent packages.

pip install azure-ai-projects

No more “why does this sample need four Azure AI packages?” moments.

(devblogs.microsoft.com)

3) Open models, but make them fast

Foundry now exposes Fireworks AI–hosted open models (preview), including DeepSeek‑V3.2, gpt‑oss‑120b, Kimi‑K2.5, and MiniMax‑M2.5, with both pay‑per‑token and provisioned throughput options.

This matters because:

  • Latency is competitive with proprietary APIs.
  • You can deploy post‑trained or fine‑tuned weights without standing up your own inference stack.
  • Pricing is explicit and predictable for production workloads.

(techcommunity.microsoft.com)

4) Model catalog keeps expanding (carefully)

The same update confirms:

  • NVIDIA Nemotron models are now first‑class in the Foundry catalog.
  • Grok 4.2 has graduated to GA status.

This doesn’t mean “switch everything now,” but it does mean model diversity without vendor lock‑in gymnastics is improving.

(devblogs.microsoft.com)

Practical upgrade checklist for .NET teams

  1. Audit packages
    • Remove Azure.AI.Agents if present.
    • Upgrade to the Foundry .NET SDK 2.0.0.
  2. Refactor client creation
    • Centralize on AIProjectClient.
    • Treat agents as a project resource, not a sidecar.
  3. Re-evaluate inference choices
    • If cost or latency is a pain point, trial a Fireworks‑hosted open model in a non‑prod environment.
  4. Lock versions intentionally
    • This is a cleanup release, but it’s still a major version. Pin it, test it, then roll forward.

Why this matters beyond this week

This update signals a clear direction: Foundry is becoming the opinionated, stable control plane for Azure‑native AI apps, not just a model menu. For .NET engineers, that translates to:

  • Cleaner APIs
  • Fewer surprise dependencies
  • A smoother path from prototype to production

No hype—just less friction. And honestly, we could all use more of that.

Further reading

  • https://devblogs.microsoft.com/foundry/whats-new-in-microsoft-foundry-mar-2026/
  • https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/whats-new-in-foundry-labs—april-2026/4509714
  • https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/announcing-fireworks-ai-on-microsoft-foundry/4500950
  • https://techcrunch.com/2026/04/02/microsoft-takes-on-ai-rivals-with-three-new-foundational-models/