"The spread of artificial intelligence (AI) agents and quantum computing has created the need for a new network infrastructure. The era will shift to a 'cognitive internet' where AI agents can understand each other's intentions, cooperate, and reason together."
Bijoy Pandey, senior vice president and general manager of Cisco Outshift, stated accordingly at the "Cisco Connect 2026 Korea" media roundtable held at Coex in Samseong-dong, Gangnam-gu, Seoul, on the 8th. Outshift is Cisco's technology research and development organization.
The concept Senior Vice President Pandey emphasized most that day was the "cognitive internet." This is a next-generation network concept in which AI, people, devices, and services go beyond simple connection to understand context, intent, and knowledge and collaborate. He said, "It is not enough for each AI agent to get smarter," adding, "We need a structure in which they can understand each other's intentions, cooperate, and reason together." He went on, "To that end, we need a protocol that aligns shared intent among agents and a fabric (an operating system that tightly weaves multiple devices or systems to move as one) that maintains collective memory across the system." Cisco views this kind of open architecture layer as the foundation of the next-generation AI internet.
Vice President Pandey also pointed out the limits of the current internet architecture. He emphasized, "With today's internet, which is laid on and designed around deterministic systems that produce the same output for the same input, it will be difficult to handle AI that reasons and collaborates and the quantum computing era based on probabilities," adding, "This change is not the evolution of individual technologies but a shift that will transform how we design, operate, and trust networks."
As a project to realize this vision, he presented the open-source "AGNTCY." AGNTCY is an open infrastructure for discovery, identity, messaging, and observability among AI agents. He said Cisco first unveiled it and then donated it to the Linux Foundation, and that more than 80 corporations and institutions, including Google Cloud, Dell Technologies, Oracle, and Red Hat, are currently participating. The point is to lay the groundwork for an era when AI agents consolidate across organizational boundaries.
Quantum networking also emerged as a major topic. Vice President Pandey said, "The Cisco Quantum Lab is working to build a full stack for distributed quantum computing that spans hardware, protocols, and applications," adding, "We are also conducting experiments on city-scale Quantum Entanglement exchange over commercial fiber." He stressed, "Quantum networking is moving beyond the lab and onto real infrastructure."
He also noted that the transition to Quantum-resistant Encryption has become an urgent task that can no longer be delayed. "As quantum computing advances more rapidly, 'Q-day,' which threatens the existing internet cryptosystem, could arrive sooner than expected, and the 'harvest now, decrypt later' threat—stealing data now and decrypting it later with quantum computers—is becoming a reality," he said. Accordingly, corporations should go beyond merely patching existing systems and build quantum-resistant security in from the product and platform stages.
He ultimately identified the network as the key to infrastructure competitiveness in the AI era. He explained, "As AI agents are trained in data centers, deployed across multiple applications, and then interact with users in real time, network traffic will remain consistently high rather than spiking temporarily." Fast graphics processing units (GPUs) alone are not enough; a network architecture that can move data without bottlenecks must be in place as well.
Cisco also introduced its "AI connectivity" strategy that day. "Scale-out" for expansion inside AI data centers and "scale-across" for consolidation between data centers are the two pillars. The scale-across strategy is based on the "Silicon One P200," a network chip Cisco designed to handle 51.2T-class large-scale AI traffic without bottlenecks. Cisco said this can process 20 billion packets per second while cutting power consumption by 65%.
Vice President Pandey said, "By leading with Silicon One-based networking and optics, we will capture the bandwidth, power efficiency, and reliability required for AI training and inference," adding, "In the AI era, the core competitiveness of networks depends on how fast, how reliably, and how safely they move data."