Build 2025 – the ultimate developer fest in the AI era

Up frontThis year’s Microsoft Build (19 – 22 May) revolved around the idea of an Agentic Web—more than 50 announcements focused on making AI agents truly operational. From Windows and Azure to GitHub and Microsoft 365 Copilot, every product tried to answer one question: How do we orchestrate models, data and workflows into self-running agents? Feel free to repurpose or trim the summary below for a newsletter, blog post or social update.

Key highlights at a glance

• Copilot Tuning & multi-agent orchestration – Fine-tune models with your own data and let agents collaborate through the new Agent2Agent protocol.

• Windows AI Foundry – Packs an AI runtime, model library (local + cloud) and secure sandbox directly into Windows, turning every PC into an inference node.

• Cross-Device Resume – Windows 11 adds a macOS-style “handoff,” so a Spotify session started on your phone can continue on the PC with one click.

• GitHub Copilot → Coding Agent – Graduates from autocomplete to an agent that can run tests, craft PR descriptions and trigger CI/CD pipelines.

• Entra Agent ID & Zero Trust – Gives AI agents a verifiable identity, centralising access control and audit.

• WSL open-sourced + dev-tool upgrades – WSL moves to GitHub; Terminal, Winget and PowerToys all level-up; the Store waives its registration fee.

• SQL Server 2025 & a unified data plane – Native RAG, vector search and GPU acceleration bring “data-as-inference” to the classic database.

Deep dive

1. Copilot platform – from “assistant” to “agent”

• Copilot Tuning (enterprise preview in June) lets you LoRA-fine-tune models on internal knowledge, with outputs gated by original permissions.

• Multi-agent collaboration & Agent Store – The Agent2Agent protocol plus Model Context Protocol (MCP) allow agents to discover each other and negotiate roles; Microsoft- and third-party agents will appear in a brand-new Agent Store.

Why it matters: Organisations can break problems into smaller, governable agents instead of betting on one giant model.

2. Windows – your “desktop AI data centre”

• Windows AI Foundry streamlines model retrieval, quantisation, deployment and inference; pull from Azure AI Foundry’s 1,900 + models or fine-tune the built-in SLM “Phi Silica.”

• Native MCP support: Future Windows apps can expose “skills” for local agents to invoke; a declarative manifest is all it takes.

• WSL open-source and new security primitives (VBS Enclave SDK, post-quantum crypto) position Windows as a one-stop AI development-and-runtime box.

3. Toolchain – DevOps meets Agentic DevOps

• GitHub Copilot is rebranded a Coding Agent: it writes tests, patches vuln code, opens PRs and even authors CI scripts.

• Visual Studio / VS Code integrate the Microsoft 365 Agents Toolkit and Agents SDK for one-click create–debug–publish.

• .NET 9, Windows Terminal and Winget updates round off the performance and productivity stack for AI-native apps.

4. Data & cloud – an AI-ready tech stack

• SQL Server 2025 public preview: vector indices, GPU-accelerated storage & retrieval—your data warehouse is now a RAG backend.

• Azure AI Foundry becomes a “model app store,” with BYOM support; two-way sync with Windows AI Foundry closes the loop from cloud selection to edge deployment.

5. Security & compliance – passports for agents

• Entra Agent ID extends Zero Trust identity management to non-human principals; paired with guardrails like MCP safety policies and auto-compliance checks, it tackles “more agents, more holes.”

Real-world demos

• Healthcare: Stanford Medicine accelerates cancer-treatment design via multi-agent workflows.

• Meteorology: The UK Met Office migrates super-computing workloads to Azure and boosts forecast accuracy by 20 percent.

• Education & accessibility: Use-cases from Peruvian teachers to AISIN hearing-aid solutions show AI’s inclusive reach.

Read More