Satya Nadella, Chairman and Chief Executive Officer of Microsoft, presents during the opening day of Microsoft Build on May 19, 2025
News

Microsoft Build 2025: Microsoft Pins Its Future on AI Agents

5 minute read
David Barry avatar
By
SAVED
Microsoft Build 2025 offered more than just product launches — it was a statement of intent. Microsoft is positioning itself as the backbone of enterprise AI.

Microsoft CEO Satya Nadella's Microsoft Build keynote described the company's vision for an AI-driven future of work.

Microsoft Build 2025 offered more than just product launches — it was a statement of intent. Microsoft is positioning itself as the backbone of enterprise AI, offering tools for everyone from solo developers to massive global brands.

A 'Rock and Roll' Glitch

Build included big product reveals, new strategies and a moment when all the warnings about putting too much faith in AI seemed justified. So let’s start there.

During a session on security, Microsoft's AI security chief Neta Haiby accidentally shared a confidential Teams chat in front of a live audience, according to CNBC.

The chat revealed private details about Microsoft’s work with Walmart, including a message stating the retail giant was “ready to ROCK AND ROLL” with Microsoft’s Entra Web and AI Gateway, a suite of integrated solutions designed to enhance secure access to internet resources and effectively manage generative AI APIs.

The message also referred to a Walmart-built AI tool called MyAssistant, which Walmart unveiled in 2023, an AI-powered search function for employees that works across iOS, Android and the company's website.

Built using Walmart proprietary data, technology and large language models in Azure OpenAI Service, the new capability is designed to understand the context of a customer’s query and generate personalized responses. The problem is that even Walmart was having issues with it, as the slide that appeared on screen noted that the tool was “overly powerful and needs guardrails.”

The slip-up happened when protesters disrupted the session and put Haiby under pressure, leading her to switch screens and share client details in the process. 

Few will miss the irony that it happened when Microsoft’s security chief was talking about ... well, security. However, it was a small demonstration that mistakes happen, and no amount of security can protect companies from human error, regardless of how tight your AI tools are.

Microsoft 365 Copilot Updates

Digital workplace professionals should take note of the announced upgrades to Microsoft 365 Copilot, aimed at providing more control, flexibility and intelligence in how organizations use AI at work.

Copilot Tuning

The headline feature is Copilot Tuning, which allows businesses to build customized AI agents using their own internal data, workflows and domain knowledge, without needing to involve data scientists. Copilot Tuning happens in Copilot Studio, Microsoft’s low-code platform for building, managing and deploying AI agents.

Jared Spataro, chief marketing officer for AI at work, shared a few scenarios for the new feature in a blog post. A law firm could use Tuning to create an agent that understands how its team drafts legal documents, including the preferred tone, formatting and structure of arguments — trained entirely on the firm’s historical files and client data.

Microsoft 365 Copilot Tuning
Microsoft

A consulting firm might fine-tune an agent to produce tailored reports for specific industries such as healthcare, retail or finance, reflecting the unique language and insights required in each sector.

All of this happens, he explained, securely within the Microsoft 365 environment. Microsoft emphasized its commitment to data protection and compliance, stating that it does not use any business data to train its generative AI models, which is especially important for industries with strict confidentiality requirements.

Multi-Agent Orchestration

The company also announced multi-agent orchestration, to coordinate and troubleshoot multiple AI agents working together on complex tasks.

orchestrating AI agents from internal and external sources
Microsoft

Instead of trying to build a single all-knowing agent, companies can now create multiple specialized agents, each with its own role, wrote Lili Cheng, corporate vice president for Microsoft Copilot Studio. 

She used employee onboarding as an example, where an HR agent manages a new hire's paperwork, an IT agent sets up accounts and devices, and a marketing agent prepares a welcome email campaign for new hires. 

Multi-agent orchestration in theory would automatically coordinate all of these tasks, with each agent handing off information to the next as needed.

sample view of choosing AI agents to connect in a single workflow
Microsoft

Microsoft is clearly now framing Copilot as more than a smart assistant, but rather a platform for building entire fleets of intelligent, business-specific agents that know your data, speak your language and improve with each use.

Build Your New Digital Coworkers

Copilot Studio is the center of all of this activity. The company boasted Copilot Studio is already present in over 230,000 organizations, including 90% of the Fortune 500. Spataro predicted there will be 1.3 billion AI agents in businesses by 2028.

With that future in mind, the company introduced Microsoft Entra Agent ID. The release uses the existing Microsoft Entra identity and access management system used for human employees and extends it to agents, giving IT departments more insight and control over who (or what) can access sensitive data and systems.

Each agent receives a unique identifier, much as employees have in Entra.

Data protection improvements are also on the way as Microsoft will soon allow Purview Information Protection to work with AI agents, so they can classify and protect sensitive data automatically.

Copilot AI Agents for SharePoint and WhatsApp

Microsoft announced custom AI agents can soon be deployed directly into SharePoint (available already) and WhatsApp (coming July 2025). That means internal teams and external customers can interact with an organization’s AI bots wherever they already work and chat. They’ve also added new tools for developers, including:

  • A Copilot Studio SDK and Agents Toolkit.
  • Easier deployment to Azure.
  • Flexible support for both pro-code and low-code solutions.
  • More publishing channels, including Microsoft Teams and custom websites.

Azure AI Foundry and Multi-Agent Workflows

Microsoft also rolled out upgrades to Azure AI Foundry, a platform for building and managing AI apps at scale. One of the biggest updates: the general release of the Azure AI Foundry Agent Service, which supports multi-agent workflows.

Instead of trying to cram everything into one complex AI model, Microsoft is encouraging developers to split tasks across specialized agents to make maintenance and dependability easier to maintain.

Learning Opportunities

Microsoft also announced support for open protocols like Agent2Agent (A2A) and Model Context Protocol (MCP) — which allow agents to talk to each other and access external data more smoothly.

The Big Picture

Whether you’re building your own digital coworker, revamping your workflows with agents, or just trying to make your Windows desktop a bit smarter, Microsoft is rolling out the red carpet for the AI-powered future.

With all this on the table, AI is more than just an addition to the workplace for Microsoft; it is central to Microsoft’s vision for its own future.

Editor's Note: Catch up on more news from the workplace vendor world:

About the Author
David Barry

David is a European-based journalist of 35 years who has spent the last 15 following the development of workplace technologies, from the early days of document management, enterprise content management and content services. Now, with the development of new remote and hybrid work models, he covers the evolution of technologies that enable collaboration, communications and work and has recently spent a great deal of time exploring the far reaches of AI, generative AI and General AI.

Main image: Dan DeLong | Microsoft
Featured Research