What happened at Microsoft Build 2025?

Sebastian Sieber
1 day ago
7 min read

#News #Event #Copilot #PowerPlatform #AI

The annual Microsoft Build event primarily centers on development, empowering developers, and embracing the future of customizations. It wouldn't be typical of Microsoft today if Copilot and agents didn't quickly become the center of attention.

Among all the announcements, new abbreviations, and features, I've compiled everything you need to know to be ready for the upcoming features.

Disclaimer: Most of the shown features are preview, experimental or only a vision of a future implementation. They are suspect to change and should not be used in a productive scenario.

Agents and Copilot

Train Copilot with Microsoft 365 Copilot Tuning

Agents built with Microsoft 365 Copilot’s Agent Builder can take advantage of Copilot Tuning, that ensures access to a tuned model is restricted to permissions of the underlying training data within an enterprise.

Copilot Tuning or training agents will improve the correctness and quality of answers based on your business data.

Model Context Protocol Server for Dynamics ERP and Dataverse

A Model Context Protocol Server (MCP) helps to enable agents to streamline executions of actions, knowledge sources or a consistent fetch of data.

Agents, MCP server, and Microsoft Dynamics 365 interconnected. Blue arrows indicate communication. Text includes Microsoft Copilot Studio, Business applications.

Microsoft has announced two default MCP Servers for Dynamics / ERP and Dataverse. Once connected, the Dataverse MCP server enables four key capabilities:

Query: Discover available tables, explore schema, and retrieve real-time data via structured or natural language queries
Knowledge and search: Let agents chat over your data, search knowledge sources, and deliver contextual answers without brittle configurations
Upload (Create/Update Records): Insert new records or update existing ones in Dataverse, with schema-aware mapping to maintain data integrity
Generate with grounding prompts: Run custom prompts grounded in real business context (e.g. summarizing a record, evaluating sentiment, or drafting a tailored response)

What's new with Copilot Studio?

Custom Prompts

Specify instructions you want to be executed by agents per data source. Also publish them among different agents.

Bring your own Model

Decide which Model to use for specific tools. All Models are based on the AI Foundry and support different brands like Llama, Stable Diffusion or Mistral

Improved File Upload for Knowledge sources

Set up files in structured folders and sections to improve the grounding for the agent

Additional Knowledge Types

Featured: Files, OneDrive, Dataverse, Salesforce, ServiceNow, Dynamics365, Azure AI Search, 1400+ connecters
Advanced: Confluence, Azure SQL, Zendesk, SharePoint List, Snowflage, Databricks, SAP (HANA), Oracle

Support for multiple agent systems

Agents can call other agents to handled specific user queries, what expands the horizon of tasks and user experience

Computer Use Tool

Allow the system to interact with any system using UI automation, without the need for APIs for e.g. connect legacy applications to automation flows.

Code interpreter to run Python code

Run calculations, advanced analytics or generating data visualizations with Python code for mission critical processes.

Python code for a temperature trend chart is shown on the left, with a line graph and analysis on temperature data displayed on the right.

Deploy Agents to SharePoint Pages

Deploy agents directly to SharePoint as a new Channel option

Copilot Studio interface showing a SharePoint deployment status. Contains navigation options like Telephony, Facebook, Skype, and more.

Test your Agents within Copilot Studio

Testing agent responses has been a vibe check only so far - with the new testing and evaluation experience you can use real world conversations or even let you generate mass data for verify answers and processing results

Development Tools and Pro Code Features

Updates and new support of the following libraries and tools

A2A protocol

As already announced a few days before Microsoft Build will the Microsoft Platform also support the Agent-to-Agent protocol. This enables Microsoft Agents to communicate with Agents build on other platforms.

Copilot enhancements and updates

Several improvements are scheduled for Microsoft 365 Copilot in general

Copilot Search and Memory

With enhanced Search and Memory capabilities answers should be return a higher quality and be more personalized

Researcher and Analyst Agents in preview

Two 1st party agents are now available in Copilot. The Researcher Agent supports deeper analysis, similar to Perplexity while the Analyst Agent helps - surprise - to analyze data records and data sets - both agents support reasoning.

OpenAI GPT 4o image generation

Do you remember the two weeks where the internet turned into a place of Studio Ghibli illustrations? This powerful image generation will be also available through Microsoft Copilot

Microsoft Discovery

A new enterprise agentic platform to bring the power of AI to scientists and engineers for more intense research

Security - Entra Agent ID

Organize AI agents in a comprehensive identity and manage access for non-human identities. See all AI Agents created using Copilot Studio and Azure AI Foundry as Enterprise Application and review their access within your tenant.

Flowchart for Microsoft Entra Agent ID showing stages: Authentication, Authorization, Identity Protection, Access Governance, Visibility.

Microsoft 365

GitHub app for Teams

The enhanced GitHub app for Teams is now faster, more intuitive and packed with new features, such as

notification cards, streamlined actions with slash commands and organized pull request

conversations with threading support

Outlook Improvements

Making it easier to navigate inboxes, by providing summaries of search results and attached files directly within emails.
Helping users prepare for meetings more quickly and effectively, by surfacing and summarizing relevant context, tasks, documents and other resources related to the meeting topic.
Create custom instructions for new drafting new emails (available today) Go to Outlook -> Settings (gearbox icon, right upper corner), select Copilot and enable custom instructions:

Outlook draft settings screen with dark theme; left sidebar shows folders. Draft instructions are focused on a direct, factual tone.

Development

Agentic DevOps

Agentic workflows with GitHub Copilot drive the vision of an agentic DevOps. Where intelligent agents collaborate with you and with each other. The agents help to automate and optimize every stage of the software lifecycle - handling bug fixes, small features, documentation, and more.

Coding agent

The Coding agent can take over tasks and proceeds with code changes based on past code implementation and internal guidelines

A new GitHub agent to extend the development experience and support devs with

offload your tasks
build new features
author tests
generate docs

SRE Agent

SRE Agent can help make your infrastructure more secure, resilient and scalable and can help detect and respond to incidents more quickly by automatic incident management

GitHub Copilot Agent mode for Visual Studio

GitHub Copilot agent mode is an advanced AI-assisted coding feature designed to act as an autonomous pair programmer. The agent takes high-level requirements for a coding task, creates a plan to complete the task and implements the work needed, iterating until the task is complete.

Foundry Local

AI with Foundry Local for Windows and macOS devices is now in preview. This new extension allows you to build your AI-powered apps completely without the necessity of an Internet connection on your machine.

This makes developing personal apps or creating agents for larger organizations with high security standards much more attractive.

Diagram of AI Foundry with components: CLI, Inference SDK, Mgmt SDK. Shows integration with Windows and Azure, plus availability on macOS.

Azure AI Agent service is now General Available

AI Foundry Agent Service connects the core pieces of Azure AI Foundry such as models, tools, and frameworks into a single runtime. It manages threads, orchestrates tool calls, enforces content safety, and integrates with identity, networking, and observability systems to ensure agents are secure, scalable, and production ready.

Diagram of Azure AI Foundry showing sections: Foundry Observability, Agent Service, Models, Knowledge & Tools, and Integrations.

Power Apps

Agent Feed

A new Agent Feed to maintain and organize your accessible and personal Agents, including a schedule and overview of results required reviews. Your personal check-in for escalation, approvals and reviews with the extension to create own personal actions.

Plan Designer Updates

The already existing Plan Designer received plenty of updates. The experience is now split in multiple agents on creating the plan. And - finally - It supports existing tables for creating your next application

Generate PCF controls with Natural Language

This is maybe one if not that highlight for all Low code / No code enthusiasts: the App Agent. Generating a new Page (PCF Control), just by providing instructions by natural language.

You can even include pictures and self drawn mockups!

The agent generates real code, based on React and Typescript, within the boundaries of the Power Platform and DLPs - vibe coding with extras.

The generated outcome is fully functionable, including clicking, hovering and search functionality. it can be further customized with the chat interface and also restored to an earlier version.

The code view is currently read only, the plan is to enable this for edits and check ins into a Repo by summer.

Power BI: Translytical task flows

Powered by Fabric User date functions, Translytical task flows enable users to automate tasks, such as updating records, dynamics notifications or triggering workflows across other systems.

(the modification of data trough is only available using SQL Databases in Fabric)

Dashboard showing "Contoso Opportunities" with revenue stats, filters for high risk, and step-by-step discount request instructions.

Windows

MCP for Windows

This is a bit surprising to me but Windows Copilot also receives a native MCP Support for Windows applications - connect / disconnect to MCP Server of installed software applications such as Figma

Conclusion

As we've come to expect from past events, Copilot and Agents are stealing the spotlight and eating up Microsoft's investment funds like it's an all-you-can-eat buffet. And why not? They're pouring money into this like it's a bottomless coffee cup.

Now, we have to wait and see if these AI-powered extensions are the real deal or just a flashy sideshow. But hey, I'm feeling optimistic! Sure, a little more love for Pro Dev support wouldn't hurt, but Microsoft is all about that low code/no code life right now.

I'm a fan of features like Agent mode and the Page building agent (App Agent), which can whip up a UI/UX faster than you can say "user-friendly." Huge for accessibility! Plus, they're embracing open source protocols like A2A and MCP.

But let's be real, we've got a long road trip ahead before we're cruising with mission-critical apps in production. AI is still figuring out CRUD operations like a toddler with a Rubik's cube.

We're making progress, learning the ropes with AI day by day, step by step. The tools are getting sleeker, and the announcements are paving the way for a brighter future for Microsoft Copilot. It's like watching a caterpillar slowly transform into a butterfly—one day, it'll be magnificent!