Microsoft Build opening keynote | KEY010

Build Conference 2025 Summary

Key Concepts

• Agentic Web: A platform shift towards applications powered by AI agents that can perform tasks autonomously or in collaboration with humans.
• Copilot: Microsoft's AI-powered assistant integrated into various development tools and platforms.
• Copilot Studio: A low-code platform for building custom agents that can connect to Copilot and Teams.
• Copilot Tuning: A feature that allows enterprises to fine-tune Copilot models with their own data, workflows, and style.
• Foundry: A comprehensive app platform for building and deploying AI-powered applications.
• Multi-Agent Orchestration: A framework for coordinating multiple agents to work together on complex tasks.
• MCP (Managed Configuration Protocol) Servers: Servers that provide configuration and data to agents.
• AI Foundry: A production line for intelligence, encompassing evals, orchestration, and RAG (Retrieval-Augmented Generation).
• Model Router: A feature that automatically selects the best OpenAI model for a given task.
• Grok: xAI's family of models, including reasoning, deep search, and response capabilities.

Introduction

Satya Nadella welcomes attendees to Build 2025, highlighting the current platform shift towards an open agentic web. The focus is on empowering developers with tools and platforms to build applications and agents that can improve the lives of individuals and organizations.

Developer Tools Updates

• Visual Studio: .NET 10 support, live preview at design time, improved Git tooling, and a new debugger for cross-platform apps. Moving to a monthly cadence for stable releases.
• VS Code: 100th release with improved multi-window support and easier staging.
• GitHub: Continued momentum in the enterprise with a focus on trust, security, compliance, auditability, and data residency.

Open Sourcing Copilot in VS Code

• GitHub Copilot has been used by more than 15 million developers.
• AI-powered capabilities are being integrated directly into the core of VS Code, open-sourcing the technology.
• Copilot has evolved from code completions to Chat, multi-file edits, and now Agents.

Agentic Capabilities and Examples

• App Modernization: Copilot can upgrade frameworks (e.g., Java 8 to Java 21, .NET 6 to .NET 9) and migrate on-premise apps to the cloud.
• Autonomous SRE Agent: Automatically triages, root-causes, and mitigates live site issues, logging incident management reports as GitHub issues.
• Full Coding Agent: Integrated into GitHub, allowing developers to assign issues (bug fixes, new features, code maintenance) to Copilot, which completes them autonomously.

Demonstration of Coding Agent

• Satya Nadella demonstrates assigning a bug fix (adding a filter for user group size on a community page) to Copilot in GitHub.
• Copilot creates a branch, starts GitHub Actions, generates a virtual machine, and commits a draft PR.
• The agent respects security measures and works in its own branch, using MCP servers configured by the developer.

OpenAI Partnership and Codex Agent

• Sam Altman (OpenAI) discusses the evolution of Codex and the vision for agentic coding experiences.
• Codex integrates deeply with GitHub, allowing developers to delegate software engineering tasks to a virtual teammate.
• Altman emphasizes the importance of adapting to the rapid rate of change in AI and leaning into new tools and workflows.
• Models will become simpler to use, more reliable, and offer features like multimodality and better tool integration.

Microsoft 365 Copilot Updates

• Microsoft 365 Copilot is now generally available, bringing together Chat, Search, Notebooks, Create, and Agents.
• Chat: Grounded on web data and work data.
• Search: Works across applications like Confluence, Google Drive, JIRA, and ServiceNow.
• Notebooks: Allows creating heterogeneous collections of data (chats, pages, documents, emails).
• Create: Turns PowerPoints into explainer videos or generates images.
• Researcher Agent: Synthesizes information from web and enterprise sources, applying deep chain-of-thought reasoning.
• Analyst Agent: Extracts insights, forecasts, and visualizations from raw data across multiple source files.
• Teams AI library supports MCP and enables A2A (Agent-to-Agent) communication with one line of code.

Agent Store

• Developers can publish agents to the Agent Store, making them discoverable and distributable across Copilot and Teams.
• Partners like Workday, ServiceNow, and LSEG have built agents that connect to Copilot and Teams.

Copilot Studio Enhancements

• Copilot Studio includes a full Kua agent, MCP, and Agent flows.
• Multi-Agent Orchestration allows building complex workflows by coordinating multiple agents (e.g., onboarding a new hire).
• Over one million agents have been built that connect with Copilot and Teams in the past year.

Copilot Tuning

• A new class of enterprise-grade agents can be built using models fine-tuned on a company's data, workflows, and style.
• Copilot can learn a company's unique tone and language.
• Customized models inherit the permissions of the source control and can be deployed to authorized users.

Demonstration of Copilot Ecosystem

• Miti Joshi demonstrates the new M365 Copilot app, Copilot Studio, and Copilot Tuning.
• The M365 Copilot app allows reasoning over a GitHub backlog using the Researcher agent.
• Copilot Studio is used to build an RFP response agent that generates proposals and posts them in Teams.
• Multi-agent Orchestration is used to connect the RFP agent to a compliance review agent.
• Copilot Tuning is used to fine-tune a model for contract writing.

Foundry as an App Platform

• Foundry is a comprehensive app platform for building and deploying AI-powered applications.
• It supports 1,900 models, including the latest from OpenAI.
• Azure OpenAI offers enterprise guarantees like high reliability, cost controls, security, compliance, and safety.
• The new model router automatically chooses the best OpenAI model for the job.

Grok from xAI on Azure

• Grok 3 from xAI is coming to Azure, offering reasoning, deep search, and response models in one.
• Elon Musk discusses the vision for Grok, emphasizing reasoning from first principles and aspiring to truth with minimal error.
• Grok 3.5 focuses on applying physics tools across all lines of reasoning.
• Musk emphasizes the importance of honesty and feedback from the developer community.

Conclusion

Satya Nadella concludes by emphasizing the importance of building stateful, multi-model, and multi-agent applications. The focus is on providing developers with the tools and platforms they need to create innovative solutions that leverage the power of AI.