Microsoft is supporting open standards to improve collaboration among AI agents from different companies.

Chief Technology Officer (CTO) Kevin Scott announced this on May 18, 2025 ahead of the company’s Build conference in Seattle.

Scott endorsed the Model Context Protocol (MCP), an open-source initiative led by Anthropic and supported by Google.

Scott believes that open protocols like MCP will boost innovation by enabling more companies to build interoperable AI systems.

Microsoft is also developing a method called structured retrieval augmentation to help AI agents retain key user information over time.

The Build conference, which begins on May 19, 2025 is expected to showcase Microsoft’s latest AI tools for developers.

🔗 Source: Reuters

🧠 Food for thought

1️⃣ AI collaboration echoes the internet’s founding principles

Microsoft’s push for an “agentic web” through standards like Model Context Protocol mirrors the approach that birthed the modern internet.

This vision closely parallels J.C.R. Licklider’s pioneering work in the late 1950s, when he proposed “man-machine symbiosis,” a collaborative relationship between humans and computers rather than machines simply mimicking human tasks 1.

Just as Licklider’s vision eventually led to ARPANET in 1969, which connected dispersed teams and machines, Microsoft’s backing of open protocols aims to create a similar interconnected ecosystem for AI agents 1.

The comparison to how hypertext protocols helped spread the internet in the 1990s represents attempts to establish universal standards that enable previously siloed systems to work together seamlessly.

This historical pattern suggests that open standards like MCP could be transformative for AI collaboration, potentially democratizing access to advanced AI capabilities beyond a handful of tech giants.

2️⃣ AI memory challenges require innovative approaches beyond brute force

Microsoft’s focus on “structured retrieval augmentation” for AI memory directly addresses a fundamental challenge in the field, how to efficiently maintain context without prohibitive computational costs.