AI agents can talk to each other — they just can't think together yet

Cisco's Outshift is developing a new architectural approach, the Internet of Cognition, to address the challenge of AI agents struggling to understand each other's intent. Current protocols enable message exchange but lack shared context, leading to inefficient coordination and isolated learning. This hinders multi-agent systems from collaborative problem-solving, as agents don't reason together about overarching objectives. For example, in a medical scenario, agents might fail to consider the complete patient needs. Outshift's solution focuses on shared context and intent, beyond just connectivity. The Internet of Cognition proposes three crucial layers for semantic collaboration among agents. These layers include Cognition State Protocols, the Cognition Fabric, and Cognition Engines to connect agents. These components focus on shared reasoning, facilitating not only communication but also the compounding of knowledge and insights. The goal is to move from agent communication to true collaboration, fostering an environment where agents can leverage each other's learnings. Outshift is seeking industry-wide collaboration on this open, enterprise-grade agentic system. The approach promises to enable agents to work toward common goals, leading to more efficient and effective utilization of AI. Outshift aims to show this approach in action via a soon to be released demo. The project aims to make AI agents work in a more connected fashion, allowing for more human-like collaboration.