Microsoft Shares Vision For AI Agents, And Backs Anthropic’s Open-Source Protocol

Microsoft’s Chief Technology Officer, Kevin Scott, announced on Sunday that the company expects artificial intelligence agents to interact and enhance memory capabilities in order to provide better service, including by integrating technologies from different companies. The tech giant also backed Anthropic’s open-source protocol.

According to Reuters, Scott shared a few updates and thoughts with the media and analysts at the headquarters in Washington, ahead of Microsoft’s Build conference in Seattle—which takes place today, and where experts will share with developers the latest tools and programs for AI systems.

Scott explained that Microsoft is focusing on encouraging the industry to adopt standards for AI and allow multiple systems, from different firms, to collaborate. The CTO also said that the tech giant is backing Anthropic and Google’s Model Context Protocol (MCP), and compared the protocol to the hypertexts linking websites in the 90s.

Anthropic also recently announced “Integrations,” a feature to connect its chatbot Claude to popular tools and apps through its MCP program.

According to Scott, the MCP has the potential to build an “agentic web” and improve collaboration. “It means that your imagination gets to drive what the agentic web becomes, not just a handful of companies that happen to see some of these problems first,” said Scott.

Microsoft also expects AI agents to develop improved memories so that the AI systems can be more helpful to users. “Most of what we’re building feels very transactional,” added Scott regarding the current agentic AI technology and acknowledging its potential.

The problem with enhancing AI models is the energy cost, as it requires more computing power. Microsoft is working on a new approach called “structured retrieval augmentation,” which allows AI agents to build a roadmap to remember conversations by extracting short bits of information.

A few days ago, Microsoft introduced a new feature for AI agents on Copilot Studio called “computer use” which allows advanced AI systems to perform complex tasks and take control of the user’s desktop and interact with applications.