2 min Applications

Microsoft enables collaboration between AI agents via new protocol

Microsoft enables collaboration between AI agents via new protocol

Microsoft envisions a future in which AI agents from different companies can collaborate and better remember their previous interactions.

This was announced by the company’s chief technology officer, Kevin Scott, on Sunday, ahead of Microsoft’s annual developer conference. Reuters reports on this.

Microsoft is expected to present new tools for developers who build AI systems at the Build conference, which will take place in Seattle on May 19.

At the company’s headquarters in Redmond, Washington, Scott said Microsoft is committed to promoting standards within the technology industry. This should enable AI agents from different manufacturers to work together. These agents are systems that can independently perform specific tasks, such as detecting and fixing software errors.

Standardized communication protocol

According to Scott, Microsoft supports the Model Context Protocol ( MCP). This standardized communication protocol allows AI applications to communicate seamlessly with external data sources and tools. It provides AI models with a universal way to access and use information beyond their own training data, leading to more robust and capable AI systems.

Scott described MCP as a technology that could form a network of collaborating AI agents. He compared it to the way hypertext grew the internet in the 1990s.

According to Scott, this offers the opportunity to contribute your own ideas to what that network becomes, rather than just a few large companies determining the direction. He also said that Microsoft is working on ways to help AI agents better remember what users have asked them to do. Until now, Scott believes that the interaction often feels very one-off.

Structured retrieval augmentation

However, he noted that improving the memory of AI systems is very costly because it requires additional computing power. That is why Microsoft is focusing on a new method: structured retrieval augmentation. This involves an agent extracting short fragments from each user interaction, creating an overview of what has been discussed.