For the past several decades, innovation in campus networking – voice over IP, wireless connectivity, digital telepresence, BYOD – was  driven by how people use the network. The transition to agentic AI marks a shift of unprecedented scale. As AI digital coworkers are onboarded by the thousands and millions, network infrastructure must adapt to the behaviors of agents. This shift calls for a complete re-evaluation of requirements for how enterprises deliver capacity, security, and visibility.

I’ve written before about the impact of massive AI deployments on data center networks. Now, the agentic shift is reaching the campus, branch, and enterprise edge.

The network requirements of agentic AI

Agentic introduces four fundamental shifts in what enterprise networks must deliver.

#1: Distributed control and visibility

Enterprise security has long relied on a perimeter model designed for human users. External (internet) traffic is isolated by firewalls and internal traffic is presumed to be generated by trusted, trained employees who are subject to company policies.

When autonomous agents begin to spawn tasks, invoke APIs, and access resources from anywhere inside the enterprise, the threat surface is everywhere. Visibility and enforcement must be pervasive and embedded from the core to the edge.

#2: Performance of a higher order

Human-driven networks scaled to human rhythms. Most traffic was driven by email, videoconferencing, database queries, and internet usage for 30-40% of the day. Agentic workloads run continuously, generating orders of magnitude more traffic, 100% of the time. AI agents call on internal and external data resources more voraciously, producing data patterns entirely different from the workforce they are supporting. The network must be engineered for sustained peak loads far beyond historical norms.

#3: Adaptive networking is fundamental

Network security response used to be measured in hours or days – the time it took to assess a new vulnerability and roll out a patch or policy update. Agentic tasks now appear, transact, and disappear in milliseconds. Bad actors can emerge and propagate in real time. Detection-to-response must compress to the timescale of the network itself – requiring intelligence that deploys new policy updates at machine speed, and hardware that adopts new packet-handling capabilities without physical replacement.

#4: Seamless operations by design

When network problem-solving ran at human speed, interoperating a mix-and-match network — one vendor’s firewall, another’s visibility tool, and the incumbent switch fabric in the middle – worked well.

When change outpaces human capacity to continuously integrate and tune disparate components, the network must work as a coherent system. Network engineers need to deploy a complete, functional solution and have it operational before the next threat cycle. Seamless integration is essential for agentic era speed.

These four shifts converge on one conclusion: the infrastructure that powered the enterprise of the last decade cannot support the enterprise of the next one.

Cisco’s vision for agentic AI operations

Meeting these demands requires every layer of the stack to be optimized and engineered to work together — software providing pervasive visibility and intelligent policy orchestration, purpose-built hardware for speed and scale, and the silicon handling packets at terabit speed while delivering visibility and adaptability.

This week at Cisco Live, we announced Cisco Cloud Control, a new platform for simpler agentic AI operations that unifies every domain into one operational platform — shared inventory, unified topology, and a single environment for teams and AI agents to work from.

What makes Cisco Cloud Control unique is that it’s built on an end-to-end product line of consistent hardware systems, running integrated network operating system software, and powered by Cisco Silicon One – the foundation for agentic AI operations.

Cisco Silicon One: The engine for the AI-ready campus

Cisco Silicon One is purpose-built for the performance and adaptability requirements of distributed enterprise networks in the agentic era. Three capabilities define how Silicon One is changing the game:

1. High Capacity: Silicon One delivers scalable performance for Campus Enterprise. It is architected for higher sustained throughput across the enterprise network. Deep on-chip buffering – a hallmark of the Silicon One architecture – absorbs traffic bursts and maintains forwarding performance under the continuous, unpredictable loads that agentic workloads generate. Silicon One also delivers industry-leading adaptive table allocation to dynamically balance resources and plan for capacity and enables switch level stacking to seamlessly operate as a single logical system with minimal disruption.
2. Seamless Visibility: Silicon One embeds proactive real-time telemetry, identity-aware forwarding, and deep traffic visibility directly into the silicon, giving operators the, “who, what, and where” of network traffic at every node. Optimized for Cisco Cloud Control and IOS XE, Silicon One offers a holistic view of network traffic patterns for end-to-end operational simplicity and reliability.
3. Adaptive Programmability: Silicon One’s programmable architecture allows new capabilities to be deployed in the field as requirements evolve, keeping infrastructure current and extending the value of every deployment. We’ve demonstrated this in production: new forwarding features, updated path-selection algorithms, and advanced telemetry capabilities have all been delivered to already-deployed Silicon One systems without a silicon spin. Silicon One also enables sub-second upgrades and rapid policy propagation, allowing Cisco Cloud Control to work in tandem with IOS XE to take real-time control of the network and respond as conditions evolve. This is machine-speed adaptability realized at the silicon level – closing the gap between detection and response that defines the agentic challenge.

Silicon One is also fused with security to meet the demands of the agentic-fueled threat landscape. Silicon One delivers hardware-accelerated MACsec and IPsec encryption with policy-aware transit, bolstered by post-quantum readiness to protect against Harvest-now, Decrypt-Later threats.

The foundation for what comes next

Cisco Silicon One now powers the new Cisco C9550 and C9350 Series Smart Switches for enterprise campus, along with over 60 Cisco systems – five purpose-built silicon families spanning hyperscaler, neocloud, enterprise, and service provider.

The campus network has always evolved alongside the way people work. Now agents are arriving, and they’re rewriting the requirements. The enterprises that move fastest will be those running on infrastructure designed for this moment – purpose-built silicon, integrated with a unified operating system and cloud-native orchestration, engineered to perform, adapt, and defend at the speed the agentic era demands.

Cisco Silicon One will power the enterprises that deliver agentic-scale capacity, security, and visibility.

Learn more about Cisco Silicon One.