Microsoft Build 2025: All the big AI announcements you need to know

1. Vision for an Open Agentic Web

• Summary: Microsoft is shifting from standalone apps to a platform enabling an open, scalable, agentic web, expanding opportunities for developers across all layers of the stack.
• Key Points:
  • Focus on open, scalable infrastructure.
  • Empowering developers to build across diverse layers.
  • Emphasis on trust, security, compliance, auditability, and data residency.

“We’re going from these few apps with vertically integrated stacks to more of a platform that enables this open scalable agentic web.”

2. Visual Studio Updates

• Summary: Visual Studio, a leading IDE for .NET and C++, receives significant updates to enhance developer productivity.
• Key Features:
  • Support for .NET 10.
  • Live preview at design time.
  • Improved Git tooling.
  • New debugger for cross-platform apps.
  • Additional enhancements for performance and usability.

3. Visual Studio Code (VS Code) Milestones

• Summary: VS Code celebrates its 100th open-source release, introducing new features and open-sourcing GitHub Copilot integration.
• Key Updates:
  • Improved multi-window support.
  • Simplified stage viewing within the editor.
  • Open-sourcing Copilot in VS Code, integrating AI capabilities into the core open-source repository.
  • App modernization in agent mode: Copilot upgrades frameworks (e.g., Java 8 to Java 21, .NET 6 to 9) and migrates on-premise apps to the cloud.

4. GitHub: Empowering Developers

• Summary: GitHub remains a central hub for developers, with GitHub Enterprise gaining traction and new AI-driven coding capabilities.
• Highlights:
  • GitHub Enterprise sees strong enterprise adoption.
  • GitHub Copilot evolves into a full coding agent, autonomously handling bug fixes, new features, and code maintenance.
  • Copilot tasks can be assigned via GitHub issues, transitioning from pair programmer to peer programmer.

5. Azure SRE Agent

• Summary: A new autonomous Site Reliability Engineering (SRE) agent automates incident management and integrates with GitHub.
• Capabilities:
  • Automatically triages, root-causes, and mitigates issues.
  • Log incident reports as GitHub issues with repair items.
  • Assigns repair tasks to GitHub Copilot for resolution.

6. Microsoft 365 Copilot Enhancements

• Summary: Microsoft 365 Copilot integrates chat, search, notebooks, creation tools, and agents into a unified, intuitive AI-driven interface.
• Features:
  • Chat: Grounded in web and work data for contextual responses.
  • Search: Spans applications like Confluence, Google Drive, Jira, and ServiceNow, beyond M365 data.
  • Notebooks: Supports heterogeneous data collections (chats, pages, documents, emails).
  • Create: Transforms PowerPoint into explainer videos or generates images.
  • Agents:
    • Researcher: Synthesizes web and enterprise data with chain-of-thought reasoning.
    • Analyst: Processes raw data from multiple sources (e.g., Excel) for insights, forecasts, and visualizations.

  • Agents integrate into Microsoft Teams, enabling action item assignments and workflow initiation via

    @mention

7. Teams AI Library and Multiplayer Agents

• Summary: The Teams AI Library simplifies building multiplayer agents with minimal code, supporting advanced memory and search capabilities.
• Key Features:
  • Supports MCP (Multiplayer Compute Protocol) with one-line code for A2A (Agent-to-Agent) interactions.
  • Adds episodic and semantic memory using Azure Search and a new retrieval system.
  • Developers can publish agents to the Agent Store for discovery across Copilot and Teams, reaching millions of users.

8. Copilot Tuning for Enterprise

• Summary: Copilot Tuning enables enterprises to customize Copilot with company-specific data, workflows, and tone.
• Details:
  • Fine-tunes models using a small set of references.
  • Inherits source control permissions for secure deployment.
  • Customized models are deployed to authorized users.
  • Future updates will deepen the understanding of the company’s expertise and knowledge.

9. Azure AI Foundry and Agent Foundry

• Summary: Azure AI Foundry and Agent Foundry simplify building, deploying, and managing AI agents and models, with support for open-source and hybrid environments.
• Key Announcements:
  • Agent Foundry:
    • Build declarative agents with minimal code via the portal.
    • Supports multi-agent orchestration for complex workflows.
    • Generally available, with connections to container apps, functions, and AKS (Azure Kubernetes Service).
    • Deploy open-source models in cloud or hybrid mode with ARC.
  • Model Router: Automatically selects the best OpenAI model, enabling multi-model support.
  • Grok from xAI: Now available on Azure, with unified throughput provisioning across multiple models.
  • Observability: New features to monitor AI impact, quality, safety, and cost.
  • Entra ID Integration: Agents receive identities, permissions, and access controls, with automatic directory listing.
  • Partnerships: ServiceNow and Workday integrate for automated agent provisioning via Entra.
  • Purview and Defender: Ensure end-to-end data protection and security against threats like wallet abuse or credential theft.

10. Foundry Local and Windows AI Foundry

• Summary: Foundry Local brings AI app development to edge and client devices, while Windows AI Foundry extends support across the full dev lifecycle.
• Foundry Local:
  • High-performance runtime for models and agents-as-a-service.
  • CLI for local app development, supported on Windows and Mac.
  • Access to pre-optimized open-source models for local execution.
• Windows AI Foundry:
  • Supports development across CPUs, GPUs, NPUs, and cloud environments.
  • Native MCP support with built-in servers (e.g., file systems, settings, windowing).
  • MCP registry for secure, vetted server discovery.
  • Used internally to build SDKs and features for Copilot PCs.

11. Open-Source Initiatives

• Summary: Microsoft continues its commitment to open source with WSL and NLWeb.
• Announcements:
  • WSL (Windows Subsystem for Linux): Now fully open source, with code available on GitHub.
  • NLWeb: Enables websites and APIs to become agentic applications with minimal effort.

• Summary: Cosmos DB and Fabric unify structured and semi-structured data for AI applications, with advanced query and transformation capabilities.
• Key Features:
  • Cosmos DB integrates with Foundry for conversational history and RAG (Retrieval-Augmented Generation) applications.
  • PostgreSQL queries now support direct LLM responses.
  • Cosmos DB in Fabric supports text, images, and audio alongside SQL.
  • Shortcut Transformations in OneLake: AI-driven ETL (Extract, Transform, Load) with pre-built transformations (e.g., audio-to-text, sentiment analysis, summarization).
  • Digital Twin Builder: No-code/low-code tool in Fabric for mapping physical assets to digital twins.

13. Nvidia Partnership and Supercomputing

• Summary: Microsoft and Nvidia collaborate to build the largest GB200-based supercomputer on Azure.
• Impact:
  • Scales AI capabilities for developers.
  • Supports advanced AI workloads and agentic applications.

14. Microsoft Discovery for Scientific R&D

• Summary: Microsoft Discovery, built on Foundry, accelerates scientific research with specialized agents for R&D.
• Capabilities:
  • Understands scientific nuances from public and proprietary data.
  • Supports biopharma and other industries in accelerating R&D.
  • Deployed by companies to enhance research workflows.

15. Closing Remarks

• Summary: The keynote emphasizes a systems-level, platform-driven approach to empower developers across the agentic web.
• Key Takeaways:
  • Comprehensive stack updates across development tools, AI, and infrastructure.
  • Focus on enabling developers to build secure, scalable, and innovative solutions.
  • Commitment to open-source, enterprise-grade, and scientific advancements.