Cisco Outshift Develops Internet of Cognition for AI Agents

Cisco's Outshift division is developing a new framework to enable multiple specialized AI agents to collaborate and coordinate complex tasks, moving beyond individual agent capabilities towards distributed artificial superintelligence. This initiative addresses the current limitation where AI agents, despite possessing domain-specific intelligence, struggle to work together cohesively without human intervention. Vijoy Pandey, senior vice president and general manager of Outshift, explains that the missing element is a "connective tissue" that transforms disparate agents into a unified team capable of shared intent, context, and reasoning.

This "connective tissue" is being built through a semantic layer called the "Internet of Cognition." This layer allows AI agents, even those operating in different domains with distinct knowledge bases and objectives, to communicate and collaborate effectively. Beneath this semantic layer lies a connectivity layer known as the "Internet of Agents." This foundational layer enables autonomous agents to discover each other, verify their identities, and exchange messages across various domains. Together, the Internet of Agents and the Internet of Cognition represent a significant step towards achieving distributed artificial superintelligence, according to Pandey.

The broader AI industry has historically focused on vertical scaling, leading to larger models trained on more data and utilizing greater computational power. This approach has yielded AI agents with advanced reasoning capabilities, akin to a "brain" that can perceive, reason, and act within digital environments. However, Pandey argues that to facilitate agentic problem-solving across diverse systems, companies, and platforms, the next crucial axis of scaling must be horizontal. This involves enabling multiple agents to interact and cooperate.

While multi-agent systems are already being explored in fields such as software engineering, drug discovery, and scientific simulations, their current performance has been described as "underwhelming." A study evaluating seven open-source multi-agent systems revealed failure rates ranging from 41% to as high as 87%. The development of the Internet of Cognition and the Internet of Agents by Cisco's Outshift aims to overcome these limitations by providing the necessary infrastructure for robust and coordinated multi-agent collaboration, paving the way for more sophisticated AI applications in healthcare and beyond.