Microsoft used Ignite 2025 to send a clear message: AI tooling is no longer meant to be fragmented. Azure AI Foundry is now Microsoft Foundry—a more unified, streamlined platform for building and managing AI agents. The rebrand signals Microsoft’s move toward an agent-first, enterprise-ready future with models, knowledge, and governance all in one place.

Here’s a look at what’s new, why the change is significant, and what to expect as organisations start adopting Microsoft Foundry.

Microsoft Foundry is a platform that helps you build AI agents—think of them as smart assistants that can understand questions, access information, and complete tasks automatically.

Here's a simple example: instead of manually searching through company documents to answer customer questions, you could create an AI agent that does this instantly. The agent searches, finds the right information, and provides accurate answers—all in seconds.

Think of Microsoft Foundry as having six main parts that work together:

Models are the AI "brains" that power your agents. Microsoft Foundry now offers access to over 11,000 different models. Why so many? Because different tasks need different capabilities.

Think of models like tools in a toolbox:

Some popular models now available:

Model Router is a new feature that automatically picks the best model for each task, balancing speed, quality, and cost. It's like having a smart assistant that chooses the right tool for each job.

The Agent Service is where you deploy your AI agents so people can actually use them.

Three important features:

Hosted Agents: Deploy your agent without worrying about servers or infrastructure. Microsoft handles all the technical stuff behind the scenes. Just build your agent and click deploy.

Multi-Agent Workflows: Create multiple specialized agents that work together. Here's how it works:

Example: For customer service, you might have:

Each agent is good at one thing, and they work together seamlessly.

Built-in Memory: Agents can remember previous conversations with users, making interactions feel more natural and helpful.

Foundry IQ connects your agents to different sources of information:

The clever part? It automatically checks permissions. If a user shouldn't see a document, the agent won't retrieve it for them.

Foundry Tools provides over 1,000 ready-made connections to systems like:

This means your agents can read from and write to these systems, just like a human user would.

Example: An agent could check a customer's order status in Salesforce, look up inventory in SAP, and send an update email—all automatically.

The Control Plane is like a dashboard that shows you everything about your agents:

What you can do:

Foundry Local lets agents run directly on computers or devices, even without an internet connection. This is useful for situations where data privacy is critical or internet access is limited.

Think of Agent 365 as “Azure AD for Agents.”

It provides:

This works even if the agent isn’t built on Microsoft Foundry.

The rebrand from Azure AI Foundry to Microsoft Foundry signals more than a name change—it represents Microsoft's strategic shift toward making AI agents a fundamental part of enterprise architecture. Here's what architects need to focus on:

The new Foundry Control Plane closes long-standing governance gaps. Every agent now receives a Microsoft Entra Agent ID, enabling:

This works across GitHub, Visual Studio, and Copilot Studio—not just Microsoft-native deployments.

Microsoft Foundry now offers both OpenAI GPT and Anthropic Claude models in one platform—something no other cloud provider can claim. The Model Router automatically optimizes for cost, latency, and quality across this catalogue of 11,000+ models.

Architectural implication: Design model-agnostic systems—your AI layer should let you swap models without touching application logic.

Agent 365 extends identity and management to agents built anywhere—not just Microsoft tools. This matters because:

Treat agents like applications: centrally registered, monitored, and policy-controlled.

Model Router's intelligent model selection can significantly reduce costs—early customer deployments reported up to 50% reduction in model costs—but architects need to design for cost optimization from day one:

The Control Plane's cost tracking becomes your primary tool for FinOps on AI workloads. Build cost monitoring into your architecture review process.

The new multi-agent workflows capability changes how we think about system design. Instead of monolithic agents, design for:

Traditional approaches will give way to multi-agent patterns as shown below:

This architectural pattern improves maintainability, reduces blast radius, and enables independent scaling.

With Managed Instance on Azure App Service, existing .NET apps can adopt AI incrementally.
Foundry’s 1,000+ connectors (MCP + OpenAPI) mean integration follows familiar patterns—no reinvention.

Defender + Purview + Entra Agent ID give agents the same protection as enterprise apps:

Security teams get full visibility from day one.

Microsoft Foundry marks a shift in how enterprises will design, govern, and operate AI systems. For architects, the opportunity is clear: reshape your patterns now so your organization can move faster, stay secure, and control costs as AI becomes a foundational layer - not an add-on.

If you found this useful, tap Subscribe at the bottom of the page to get future updates straight to your inbox.