Microsoft Build 2025: our 10 highlights for developers and AI teams

Microsoft Build 2025: 10 highlights that matter for developers and AI teams
Table of Contents

Microsoft Build 2025 marked a pivotal moment in the tech landscape, showcasing Microsoft’s ambitious strides in AI and developer tools. For those of us helping clients navigate Microsoft’s ecosystem, this year’s announcements were particularly exciting and full of real impact.

Here’s a breakdown of the 10 most important takeaways:

 

GitHub Copilot evolves into a peer programmer

GitHub Copilot is no longer just autocomplete for code. It’s evolving into what Microsoft calls a peer programmer. Under the project name “Padawan,” this new capability allows the AI to autonomously fix bugs, refactor code, and improve documentation in well-tested codebases. The goal is to take care of routine tasks, so developers can focus on the hard stuff.

 

Copilot Studio introduces multi-agent capabilities

You can now build multi-agent systems in Copilot Studio. These agents can delegate tasks to one another, making it possible to design workflows where AI does more of the coordination, and less just answering prompts. Tools like the Microsoft 365 Agent Builder and Azure AI Agent Service are central to this new setup.

 

NLWeb brings natural language to websites

Microsoft is betting big on conversational interfaces. A new project called NLWeb, led by R.V. Guha, introduces a framework for adding natural language input to websites. The idea? Make it easier for users to interact with services through simple, intuitive language, not clicks and forms.

 

Copilot Studio opens up for developers

Developers can now plug into Microsoft 365 Copilot APIs (the retrieval API is already in preview) and even connect custom models from Azure Foundry. This opens the door to build bespoke copilots that pull from proprietary data, models, or third-party services, fully integrated into Microsoft 365 apps.

 

Microsoft embraces the Model Context Protocol

Microsoft is adopting the Model Context Protocol (MCP) across its platforms, including GitHub, Copilot Studio, Dynamics 365, Azure, and Windows. MCP is a new standard that allows different AI agents and models to interoperate, unlocking more modular and scalable AI solutions.

 

Windows AI Foundry debuts at Microsoft Build

The Windows Copilot Runtime has a new name: Windows AI Foundry. It’s a full-stack platform for building and running AI-powered applications, both in the cloud and on-device (even on macOS). It supports everything from model optimisation to deployment, making Windows a serious AI development platform.

 

Windows Subsystem for Linux (WSL) goes open source

Microsoft is open-sourcing most of WSL, including the CLI tools and services. For developers, this means more transparency, more control, and potentially faster innovation from the community. It also reaffirms Microsoft’s ongoing support for open-source development.

 

Azure AI Foundry adds Grok models

In a surprise twist, Microsoft announced a partnership with xAI, Elon Musk’s AI company. The result: the Grok 3 and Grok 3 Mini models are now available in Azure AI Foundry. This gives Azure developers access to yet another family of powerful LLMs to build on top of.

 

Microsoft 365 Copilot gets smarter

New updates to Microsoft 365 Copilot bring better multi-agent orchestration, improved tuning, and more context awareness across apps. It can now analyse emails, files, and content more intelligently, helping users get expert-level assistance without switching tools.

 

Microsoft Discovery for scientific research

Microsoft introduced a new platform called Microsoft Discovery, built for the scientific community. It uses AI agents to assist across the full research lifecycle — from hypothesis generation to simulation and reporting. It’s early days, but the vision is clear: AI that accelerates science.

 

Our final thoughts

We’ve long believed that Microsoft’s stack is one of the most powerful platforms for real-world Enterprise AI applications. What Build 2025 confirms is that it’s also becoming more modular, more interoperable, and more agent-driven than ever before.

If you’re building with Copilot Studio, exploring Azure AI Foundry, or just trying to make sense of what these updates mean for your team, we’re here to help!

 

Stellium

May 23, 2025