For years, enterprise networking was seen as a commodity, an abstraction layer that the cloud was responsible for hiding. But the arrival of artificial intelligence has changed the rules of the game. At Cisco Live 2025, the company presented a set of innovations that place the network at the center of AI infrastructure, with a renewed focus on security, performance, and flexibility.

The Network as the New PCI Bus

Tom Gillis, senior vice president and general manager of Cisco's Infrastructure and Security Group, explains that in the AI era, the network plays the same role as the PCI bus inside a server. Distributed AI systems require memory, compute, GPU, and storage to work together across physical infrastructures at scale. The network is the backplane that makes this possible. Customers have come to see it as the only element they can trust, a radical shift from the abstraction trend of the last decade.

AI Transforms Cisco's Internal Development

The impact of AI is not only reflected in Cisco's products but also in how they are developed. Gillis leads a team of approximately 12,000 software developers, and AI coding tools have radically transformed their work. Previous generations of AI coding tools achieved significant improvements in new projects but hit a ceiling with complex legacy products, such as a Cisco Catalyst switch or firewall, which can contain between 50 and 100 million lines of code. New AI coding tools have removed that limit, accelerating development across the entire product catalog.

However, this acceleration has a dark side: vulnerability discovery. Cutting-edge AI models like Anthropic's Claude Mythos can already understand complete complex codebases and are finding vulnerabilities that humans had been unable to detect. For Gillis, "cutting-edge models are finding vulnerabilities on an unprecedented scale, and it's not a one-time thing. They will continue to find new vulnerabilities."

Security from the Linux Kernel: eBPF and Live Protect

Cisco's response to ongoing vulnerabilities is based on Isovalent, its commercial platform built on the open-source project Cilium, which uses eBPF technology integrated into the Linux kernel. eBPF allows inspecting memory, intercepting every system call, and modifying those calls without needing to reboot or modify system binaries. This enables Live Protect, a function integrated directly into Cisco's network operating systems like NX-OS and IOS. A compensating control can be scoped to a specific process ID and file, blocking a specific action without affecting anything else on the system. For an administrator, the experience is seeing a vulnerability flagged in the Nexus dashboard with a button to apply a shield.

This approach is especially relevant in a context where enterprise security must adapt to an ever-evolving threat landscape. The ability to apply hot patches without service interruption is a significant advancement.

Unifying VMs, Containers, and AI in a Single Infrastructure

Not all workloads are AI. Most enterprises still rely on virtual machines (VMs) and container-based applications. Cisco has developed a software bridge based on Isovalent that allows migrating VMs to Kubernetes without changing their IP addresses, resolving the friction point between layer 2 (VMware) and layer 3 (Kubernetes). The result is that legacy VM workloads, container-based applications, and AI workloads run together on the same infrastructure, without the need for a forced full migration.

Gillis demonstrated this vision on the main stage at Cisco Live. VM-based and Kubernetes-based workloads appear as peers in the same Nexus dashboard, while the software bridge handles the layer 2 to layer 3 translation underneath. This is especially relevant at a time when VMware under Broadcom creates market uncertainty, and companies seek flexible alternatives.

The Future of Networking: AI Agents and Task-Based Access Control

The next challenge is already visible. As AI agents begin to act on behalf of users in enterprise systems, the network faces a new access control problem for which it was not designed. A typical enterprise user has password-based access to hundreds of applications, with credentials rotated every six months. Extending that same access to an agent is too permissive. "We need to implement task-based controls, much more ephemeral, for agents," says Gillis. "An agent authorized to submit an expense report should have no ability to make purchases; I don't want the agent to buy a Porsche."

Cisco is addressing this through Cisco Secure Access, its SSE solution, and its hybrid mesh firewall, with task-defined and session-specific controls for both user-application and server-server scenarios. This approach aligns with trends in contextual enterprise agents and generative AI implementation in workflows.

Next Steps: A Unified Architecture

Beyond the event, Gillis acknowledged that Cisco has additional announcements planned for the fall. "We will make some announcements in the fall that I think will be surprising," he said, without providing details. His short-term vision is a single infrastructure architecture covering all types of applications. "Within a year, I expect customers to realize they can build an infrastructure capable of driving their AI applications—which are the future—and at the same time, their Kubernetes-based applications—which are the present—and their VM-based applications—which are the past—all with a single design, a single architecture."

This vision of unification contrasts with the current fragmentation, where companies often maintain silos for different workload types. Cisco's proposal not only simplifies management but also reduces costs and improves agility, critical aspects in an environment where AI models like Nemotron 3 Ultra demand robust and flexible infrastructure.

Original source: ComputerWorld. Analysis and adaptation by ForgeNEX.