Select Language:
Microsoft Aims for AI Collaboration at Build Conference
REDMOND, WA: Microsoft envisions a future where artificial intelligence (AI) agents from various companies can collaborate seamlessly and retain better memories of their interactions. This vision was shared by the company’s Chief Technology Officer, Kevin Scott, just before the annual Build conference.
Taking place in Seattle on May 19, developers expect Microsoft to announce advancements in tools aimed at enhancing AI system development.
At a pre-conference briefing at Microsoft’s headquarters, Scott emphasized the company’s commitment to fostering industry standards that facilitate cooperation between AI agents created by different companies. These agents are designed to perform specific tasks autonomously, such as identifying and fixing software bugs.
Scott highlighted Microsoft’s support for a technology known as the Model Context Protocol (MCP), an open-source protocol backed by Google-supported Anthropic. He compared MCP’s potential to foster an "agentic web" to the pivotal hypertext protocols that ignited the internet revolution in the 1990s.
“This means the development of the agentic web can be driven by imagination, rather than just by a select group of companies addressing a problem first,” Scott remarked.
Additionally, he discussed Microsoft’s aim to enhance how AI agents remember user interactions. Currently, Scott noted, “most of what we’re building feels quite transactional.” However, improving an AI agent’s memory comes at a steep cost due to the increased computing power required.
Microsoft is investigating a novel method called structured retrieval augmentation, where an agent would capture brief excerpts of each exchange with a user, mapping out the conversation’s flow.
“This reflects a fundamental principle of how biological brains learn; you don’t need to consciously recall every single detail when tackling a problem,” Scott explained.
ChatGPT
Perplexity
Gemini AI
Grok AI
