Microsoft just taught its AI agents to talk to each other—and it could transform how we work

Microsoft has announced a significant expansion of its Copilot Studio platform in Build 2025 today, where multi -agent systems have provided the various artificial intelligence agency to cooperate in complex work tasks, along with new developers tools, safety improvements, and integration with WhatsApp.

The ads group is the most ambitious Microsoft attempt to make artificial intelligence agents more practical for institutions, addressing the main restrictions that hindered the adoption of the technology of the wider agents in business settings.

“We see from customers who are passing wide -scale production that governance and observation can become more important,” Ray Smith, Vice President of the Microsoft Intelligence Agency, said in an exclusive interview with Venturebeat. “The beauty of the Copilot Studio is that we offer an infrastructure framework with a compact life cycle management and comprehensive governance capabilities.”

How the new multi -agent microsoft system transfers the Foundation’s workflow tasks

At the heart of ads, there is a new Microsoft multi -agent system, which enables agents designer using Copilot Studio, Microsoft 365, Azure Ai Agents Service and Azure Fabric to work together, and delegate tasks to each other to complete complex business operations.

This ability addresses a fundamental challenge that organizations faced when implementing the agent technology. “The creation of a reliable operation within one agent is very difficult,” Smith explained. “Dividing it into multiple factors does not improve the ability to maintain and simplify the construction of solutions, but it also greatly enhances comprehensive reliability.”

The system allows scenarios such as COPILOT studio agent withdraws sales data from CRM, delivered to Microsoft 365 to formulate a Word proposal, then leads to the operation of another agent for follow -up in Outlook.

Microsoft also emphasizes the susceptibility to the agent protocol to the recently announced agent by Google, which may enable the platform agent’s connection.

The “computer use” feature brings a human interface that resembles the human intelligence factors without API dependencies

Another major advertisement is “Use of the computer” for Copilot Studio agents, which allows agents to interact with desktop applications and web sites by controlling the facades directly – clicking the buttons, moving in the menus, and writing in the fields – even when applications programming facades are not available.

“When applications are not available, this feature can interact directly with user interfaces – whether it is desktop applications, browsers or other platforms,” ​​Smith said. “It provides what we call” the lack of expansion “and works on the basis of intention rather than the coordinates of the pixels, unlike automating the traditional desktop. This approach directed towards the target makes it significantly more powerful.”

This possibility is currently available through the Microsoft Frontier program for qualified customers with more than 500,000 COPILOT studio messages and an environment based in the United States.

Choose customized models and high -powered analyzes

Microsoft gives institutions more flexibility with artificial intelligence models by enabling them to bring custom models from Azure Ai Foundry to Copilot Studio. This includes reaching more than 1900 styles, including the latest Openai GPT-4.1, Llama and Deepseek.

Smith said: “Start with ready -made models because they are really great and constantly improve.” “Companies usually choose to set these models when you need to integrate a specific field language, unique use cases, historical data, or customer requirements. This customization ultimately pays greater efficiency or improves accuracy.”

The company also adds a feature of translator icon that brings Python capabilities to Copilot Studio agents, enables data analysis, perception, and complex accounts without leaving the COPILOT studio environment.

Smith highlighted financial applications as a special power: “In financial analysis and services, we have seen a remarkable breakthrough during the past six months.” “Deep Thinking Models, which work with reinforcement, can selfly define any process that produces quantitative measurable outputs.”

He added that these capabilities excel in “complex financial analysis where users need to create a code to create charts, produce specific outputs, or make detailed financial assessments.”

WhatsApp integration extends to AI Agent to billions of international users

From early July, organizations will be able to publish Copilot Studio agents to WhatsApp, allowing them to reach customers through one of the world’s most correspondence platforms.

“WhatsApp clearly is a basic channel. At the global level, it is very huge,” Smith explained. “So this has become a high priority for us through different channels and integration, to lock this as a way for the final users to interact at a time that suits them.”

This addition, along with the new SharePoint channel (is now generally available), is largely working to expand the scope of dedicated agents outside the Microsoft ecosystem.

Microsoft blocks the developer experience gap with the extension

For professional developers, Microsoft launch the VISUAL StuDIO Code for Copilot Studio, bringing familiar tools and workflow to the developer of the agent. The extension provides features such as Intelliseense, Color coordination, and “searching for all references” functions, enabling developers to adjust the agents directly from within Visual Studio Code.

IT officials also gain new tools, including the Central “agents and connectors” page at the Microsoft 365 to manage the full life cycle of Copilot agents. This interface allows officials to display all agents, filtering according to descriptive data, setting allergies, managing conductor behavior, prohibiting or deleting factors that violate security policies.

Discovery Ai-Age-Agent Discovery solves “Which agent I need?” A problem for the final users

With institutions developing more specialized agents, Microsoft deals with the challenge of the detection of the agent through the recommendations of the new engineer agent. This feature suggests the most relevant agent based on the user’s needs, and a dynamic recommendation for delivery processes when necessary.

“It is mainly similar to the style of the jacket of an agent,” Smith explained, referring to the generation -centered generation. “He uses the bases and indexing data based on the description of the agent. When you describe the task, they are looking and going,” hey, I think you should look at this agent. “

The system only recommends the agents that the user can access, regarding current permissions.

Overcoming the obstacles to the implementation of the Foundation AI: Microsoft Road Map for success

Despite the new capabilities, organizations still face major challenges when implementing multi -agent systems. Smith recommends focusing on specified cases instead of trying to comprehensive transformations from the beginning.

He said: “The best customer practices who focused on the state of use or a problem to start, and shrink to parts, and looked at the inclusion of an agent in different parts of this process.” “They move from one agent to two factors, and then effectively sew together in a multi -agent solution.”

He stressed the importance of practical experience with the tools: “You need to know what is not … It is a mixture of tools that you will use to transform your business.”

The Microsoft Enterprise AI strategy is to lead the next wave of shift in the workplace

These ads collectively enhance Microsoft’s position in the Enterprise AI market, where competition with Google, Amazon and Ai continues startups specialized in intensifying them.

By addressing the requirements of the main institution such as safety, governance and inter -operation, with the expansion of the statute of the statute simultaneously through features such as computer use and the interpretation of software instructions, Microsoft creates a more complete presentation of institutions that look forward to spreading artificial intelligence agents widely.

Smith pointed out that “Microsoft is in a pioneering position in building and managing these agents or employer’s workforce.” “I think we have reached an important turning point for some of the main features that customers ask.”

The company’s double approach-supports both the non-code tools for business users and professional development environments-also distinguish it from competitors who focus mainly on developers or business users.

“Our platform serves experienced developers who prefer to use Visual Studio to direct access to applications and frameworks.” “It also accommodates business users with limited coding experience who want to start with a specific commercial problem and the initial model solutions quickly using the natural language.”

While Microsoft has dealt with many of the main challenges of adopting the AI ​​with these ads, Smith hints at future developments that focus on discovering the agent and easy access to it.

He said: “I think the next piece will be all these agents that appear inside the market or catalog and then can be accessed from artificial intelligence assistants such as Copilot or Microsoft 365 Copilot and beyond.”

Build 2025 ads are an important step forward for the Microsoft AI strategy, but the company faces the challenge of pushing the adoption of these technologies widely through industries and helping organizations achieve the value of concrete businesses from their investment in artificial intelligence.

Smith also concluded, “At the core of all transformation is not a technique in itself. It is related to providing more expensive, faster, and higher quality solutions, which ultimately creates superior customer experience.”