AI Agents & the Future at Microsoft Build 2025

Microsoft Build 2025 ushered in the age of AI agents. This year’s announcements signaled a paradigm shift; not just in how developers build applications, but in how intelligent systems will operate across organizations, users, and the internet itself. At the heart of this transformation lies Microsoft’s bold vision for an open agentic web – an internet where AI agents can intelligently and securely interact with data, services, and one another.

From innovations in Azure AI Foundry to expanded support for open standards like the Model Context Protocol (MCP), Microsoft is laying the groundwork for a future where AI agents aren’t just embedded in applications; they become the key driver of interaction and intelligence.

AI agents represent the next leap in intelligent computing. Unlike traditional bots or assistants, agents can perceive context, reason, and act independently to complete complex tasks. They can schedule meetings, analyze documents, surface customer insights, or even orchestrate workflows across departments, all while continuously learning and improving.

At Build 2025, Microsoft unveiled foundational work toward enabling an open agentic web. This vision promotes a decentralized ecosystem in which AI agents built by different organizations, using various platforms and models, can interoperate seamlessly.

Key to this effort is Microsoft’s support for the Model Context Protocol (MCP), a shared standard that enables AI agents to securely access and act upon external data and services. Microsoft is not just implementing MCP across products like GitHub Copilot, Copilot Studio, Dynamics 365, and Azure AI Foundry, it has also joined the MCP Steering Committee to help evolve the protocol.

Additionally, Microsoft introduced NLWeb, an open framework akin to HTML, but for AI agents. NLWeb allows websites to expose natural language interfaces to agents, making their content discoverable and actionable in new ways. By enabling agent-friendly web architecture, Microsoft is championing a future where users can interact with the internet through intelligent, conversational agents instead of traditional clicks and forms.

Powering this agentic future is Azure AI Foundry—Microsoft’s unified platform for building, managing, and deploying AI models and agents. At Build, several powerful upgrades were announced that elevate Azure AI Foundry into a full-stack solution for enterprise-grade agent orchestration.

Model ecosystem expansion

Azure AI Foundry now hosts over 1,900 AI models from Microsoft and partners. This includes the addition of cutting-edge models like xAI’s Grok 3 and Grok 3Mini, as well as upcoming releases like Flux Pro 1.1 and Sora. Developers can now access these models through a single platform to fine-tune, deploy, and integrate them into applications.

To streamline this process, Microsoft introduced a Model Leaderboard for comparing model performance across categories and use cases, and a Model Router that dynamically selects the optimal model for any given query in real time.

Agent-oriented architecture

One of the most transformative updates is the general availability of Azure AI Foundry Agent Service. This tool allows developers to create, coordinate, and govern fleets of specialized AI agents that can collaborate on sophisticated tasks. Agents built through the Foundry can communicate using open protocols like MCP and Agent-to-Agent (A2A), allowing them to function as modular components in large, distributed digital workforces.

Importantly, these capabilities are backed by built-in observability and governance. Azure AI Foundry provides tools to track agent performance, safety, cost, and compliance. Enterprises can manage agent identity using Microsoft Entra Agent ID, automatically assigning each agent a unique identity in the organization’s directory to avoid agent sprawl and security risks.

AI agents are only as powerful as the data they can access. Microsoft is extending Azure’s capabilities to unify structured and unstructured data, enabling agents to act on real-time insights with minimal friction.

A key announcement was the integration of Azure Cosmos DB (NoSQL) into Microsoft Fabric, allowing developers to incorporate globally scalable, operational data directly into Fabric’s unified analytics environment. This enables AI agents to draw from both real-time and warehoused data, streamlining everything from personalization to operational optimization.

Additionally, a new Azure Databricks connector for AI Foundry (in preview) allows agents to access Spark jobs and datasets directly, enhancing their ability to perform live analytics and support data-intensive workloads.

These integrations reflect a broader strategy: breaking down data silos to support intelligent decision-making at scale. Whether tapping into live telemetry or long-term data lakes, Azure is evolving into a seamlessly connected data foundation for AI.