Microsoft is moving AI agents from experimental projects into real enterprise production systems, backed by new infrastructure that gives agents secure access to company data, their own identities, and reliable context about business operations. At its Build 2026 conference, the company announced a suite of tools designed specifically for agents rather than human users, signaling a fundamental shift in how enterprises will deploy AI coding assistants and autonomous workflows.

What Are the New Agent Infrastructure Tools Microsoft Just Announced?

Microsoft introduced what it calls the "IQ" family of products, each designed to give agents different types of context and access within enterprise systems. These are not consumer-facing tools; instead, they function as headless APIs that developers connect to their agents. The company announced several new capabilities at Build:

• Foundry IQ: Provides agents with access to unstructured knowledge and enterprise data repositories, allowing them to retrieve information without requiring human intervention through traditional interfaces.
• Fabric IQ: Enables agents to interact directly with structured business data stored in Microsoft Fabric and Power BI, bypassing the need for agents to navigate through human-facing reports.
• Work IQ: Serves as the agent-facing interface for Microsoft 365 applications including Outlook, Teams, Word, and SharePoint, allowing agents to read emails, access documents, and interact with collaboration tools.
• Web IQ: Provides agents with a headless web search capability that can search the internet, process videos, and perform browsing tasks automatically at high speed.

Marco Casalaina, Microsoft's VP of Products for Core AI and AI Futurist, explained that these tools represent a fundamental rethinking of how agents access enterprise information. "All of the IQs are headless," he stated, meaning they're designed for machine-to-machine communication rather than human interaction.

Marco Casalaina, Microsoft's VP of Products for Core AI and AI Futurist

How Are Enterprises Securing Agent Access to Sensitive Data?

One of the most significant announcements involves agent identity. Microsoft is extending its Entra identity system, which the company describes as the world's largest identity system for human users, to support agents as well. This means agents can now have their own email inboxes, Teams accounts, and document access permissions, just like human employees.

"In order for Work IQ to see my email, Teams messages, documents and stuff like that, I have to be able to authenticate it on behalf of me," explained Marco Casalaina, VP Products, Core AI and AI Futurist at Microsoft.

Marco Casalaina, VP Products, Core AI and AI Futurist at Microsoft

This approach solves a critical enterprise problem: how to give agents access to sensitive business data while maintaining security and compliance. Rather than creating a single privileged account for all agents, each agent can have its own authenticated identity with specific permissions tied to its role and responsibilities. The IQs themselves are exposed as MCP servers, which are self-describing APIs with built-in authentication layers and capability declarations.

What Does This Mean for GitHub Copilot and Enterprise Development?

GitHub Copilot sits at the foundation of Microsoft's agent strategy. The company is positioning Copilot not just as a code completion tool, but as the core AI layer for building and deploying agents across enterprises. At Build, Microsoft announced several supporting capabilities designed to make agents more reliable and observable in production environments.

Casalaina described a multi-layered approach to agent infrastructure. At the base is model choice, with Microsoft offering OpenAI's GPT models, Anthropic's Claude (including the newly launched Claude Opus 4.8 on Azure), and Microsoft's own MAI models, which are optimized for token efficiency and customization. Above that sits hosted agents in Foundry, which automatically handles scaling and containerization. The Foundry control plane provides observability into agent costs, token usage, and correctness, allowing teams to run continuous evaluations and ensure agents don't drift from their intended behavior.

How to Implement Agent Governance and Optimization in Your Organization

Microsoft is introducing new tools to help enterprises manage agents at scale. These include capabilities for monitoring, evaluation, and continuous improvement:

• Agent Optimizer: A new evaluation system that allows granular assessment of whether agents are working correctly, with the ability to automatically modify prompts and agent configurations to improve performance over time.
• Continuous Evaluation: Built into the Foundry control plane, this feature enables teams to sample agent interactions, run evaluations, and verify that agents continue to work correctly without drifting from their intended behavior.
• Cost and Token Observability: The control plane provides visibility into how much each agent costs to run, how many tokens it consumes, and whether it's delivering correct results, enabling teams to optimize resource allocation.

Casalaina emphasized that enterprises need far more than just access to powerful AI models. "The winning platform will be the one that gives them reliable context, governance, identity, memory, and secure access to enterprise data," he noted. This reflects a broader industry shift away from treating AI agents as standalone tools and toward treating them as integrated components of enterprise infrastructure.

Casalaina

Why Is Model Choice Still Central to Microsoft's Strategy?

Despite building its own MAI models, Microsoft continues to emphasize that enterprises should have flexibility in choosing which models power their agents. The company offers OpenAI's GPT frontier models, Anthropic's Claude models, and its own MAI models, all accessible through Azure and Foundry. This approach reflects a recognition that different enterprises have different requirements, compliance needs, and cost constraints.

The MAI models are specifically designed for enterprise customization. Unlike general-purpose frontier models, they're built for token efficiency, optimization, and the ability to fine-tune on customer-specific datasets. This allows enterprises to build agents that are tailored to their unique business processes and data without requiring massive computational resources.

Casalaina's role as AI Futurist gives him a unique vantage point on where enterprise AI is heading. He described the position as being "the first person to try anything new" at Microsoft, constantly testing emerging capabilities across the company. His focus is on what he calls "the immediate future," roughly one year out from now, which is where he sees agents moving from experimental projects into core business infrastructure.

The announcements at Build 2026 suggest that the next phase of enterprise AI adoption will be defined not by the power of individual models, but by the infrastructure that connects agents to company data, governs their behavior, and ensures they operate securely and reliably at scale.