Microsoft wants AI 'agents' to work together and remember things

By Admin

19 May, 2025

At the forefront of the artificial intelligence revolution, Microsoft is charting a bold new course one where AI agents don’t just execute tasks independently but collaborate across companies and remember what you’ve asked them to do. This forward-looking vision was outlined by Chief Technology Officer Kevin Scott on Sunday, just ahead of the company’s annual Build conference in Seattle on May 19.

AI Agents That Work Together

Microsoft envisions a future where artificial intelligence agents from different companies can seamlessly interact, collaborate, and build on each other’s capabilities. These agents, essentially task-specific AI systems such as those that can fix software bugs or manage workflow automation could one day work hand-in-hand across organizational boundaries. The idea is to move beyond isolated AI services to a connected network of intelligent agents capable of cooperation.

To make this vision a reality, Microsoft is pushing for the widespread adoption of common industry standards. At its Redmond headquarters, Scott emphasized the importance of industry collaboration in enabling these multi-agent systems to communicate effectively, much like humans working in teams.

The Role of Model Context Protocol (MCP)

A key piece of this vision is Microsoft’s support for the Model Context Protocol (MCP) — an open-source initiative introduced by Anthropic, a Google-backed AI firm. MCP is designed to standardize the way AI agents share context, making it easier for them to understand one another’s actions and intentions.

Scott likened MCP to the hypertext protocols that enabled the internet’s rapid growth in the 1990s. He described the potential of MCP to create an “agentic web” — a connected, interoperable ecosystem of AI agents that could empower users and developers alike. “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,” he said.

Making AI Remember: Structured Retrieval Augmentation

While AI agents today often feel transactional — responding to prompts without any lasting memory — Microsoft is aiming to change that. Scott highlighted the company’s investment in helping AI systems remember past interactions more intelligently. Rather than relying on expensive memory-intensive methods, Microsoft is focusing on a technique called structured retrieval augmentation.

This approach allows AI agents to extract and store short snippets from conversations, essentially creating a mental map of what has been discussed. Scott explained, “This is a core part of how you train a biological brain — you don't brute force everything in your head every time you need to solve a particular problem.” By mimicking how humans recall past interactions, Microsoft’s AI agents could become more useful and contextually aware over time.

Looking Ahead

Despite recent headlines about layoffs affecting about 3% of Microsoft’s workforce — described by one executive as a “day with a lot of tears” — the company remains firmly focused on AI innovation. The Build conference is expected to showcase new tools for developers, especially those working on AI systems, reinforcing Microsoft's commitment to advancing the field.

Microsoft’s vision of collaborative, memory-enhanced AI agents promises to redefine how we interact with intelligent systems. With standards like MCP and innovations in AI memory management, the company is laying the groundwork for a more interconnected and capable AI-driven future.