AI agents today can be linked to work together, yet they lack the ability to truly think in unison. This gap—highlighted by Vijoy Pandey of Cisco’s Outshift—presents a stumbling block for future AI advancements. While agents may be connected via workflows or overseen by supervisor models, they often lack deep semantic alignment or shared understanding, meaning collaboration essentially starts from zero each time. Pandey envisions a next-gen infrastructure dubbed the “internet of cognition” to solve this.
Shared cognition means AI agents can combine forces to create solutions they weren’t specifically programmed for, all without human help. Drawing parallels to human evolution—from solitary intelligence to sophisticated collective cognition—he perspectives that meaningful cooperation involves shared intent, context, and innovation. Cisco’s team is developing new protocols (SSTP, LSTP, CSTP) to foster this, aiming to distribute intelligence and synchronize cognition across multiple endpoints.
On a practical front, Cisco has deployed AI agents in site reliability engineering, automating numerous workflows and reducing deployment times from hours to seconds, while cutting issue rates by 80%. Pandey stresses AI is a tool that needs to be combined with traditional programming for the best results.
For further insights, the podcast “Beyond the Pilot” explores how this technology boosts error detection, why collaboration must remain open and interoperable, and how Cisco’s open source project Agntcy supports critical AI functions.
