3 pillars to enhance networks for agentic AI infrastructure

3 principles of workplace AI

According to Jeetu Patel, president and chief product officer at Cisco, technology is undergoing one of the most consequential shifts in human history.

This shift began with the generative AI era when intelligent chatbots started answering users' questions. GenAI laid the foundation, but agentic AI still requires an optimized network infrastructure to enable the next generation of AI.

Patel outlined critical constraints that need to evolve to support agentic AI, such as power, compute and data. While all these aspects are important, Patel highlighted the network as an often-overlooked limitation.

GPUs that power AI require fast and efficient networks to enable data processing. Networking is critical to AI, but it can also be a constraint because slow networks with insufficient bandwidth can leave GPUs idle. This wastes time and money, similar to burning cash, Patel said.

"The way you can avoid idle time and get the GPU to full utilization is to make sure that the packets get there on time," he said.

Anurag Dhingra, senior vice president and general manager of enterprise connectivity and collaboration at Cisco, outlined the company's three-pillar architecture for AI, which includes the following components:

1. Simplified network operations with AI.
2. Scalable, secure and future-oriented devices.
3. Security-infused network fabric.

Simplified network operations with AI

This component includes a unified platform for network management across the enterprise. It combines Cisco's Catalyst and Meraki platforms and enables network teams to manage both systems in the cloud, Dhingra said. However, enterprises can use the platform across different deployment locations, whether on-premises, in the cloud or through a hybrid approach.

In addition, the platform provides network professionals with end-to-end visibility of their network devices, so teams can manage all devices from a single platform.

The platform also includes an AI assistant baked into the platform to help administrators manage and troubleshoot network configurations. The AI assistant will include agentic AI, which Dhingra described as a true enabler of IT administration.

Now that AI is moving to agentic ops -- an era in which DJ Sampath, senior vice president of AI software and platform at Cisco, described as a paradigm shift in how IT environments operate -- organizations must think about how to optimize network operations to support this.

"When you start to think about agents, what really becomes important is being able to operate an environment where these agents are doing the things they're doing," Sampath said. "You need a next level of operations that comes together."

Scalable, secure and future-oriented devices

Enterprise AI requires new network devices to accommodate AI workloads, Dhingra said. Cisco plans to release the following to meet those demands:

• Advanced switches. Powerful switches with silicon processors to run security workloads.
• Secure routers. New routers should have improved capabilities, such as improved routing, next-generation firewalls and software-defined WAN.
• Wi-Fi 7 APs. Cisco plans to release a more comprehensive Wi-Fi 7 portfolio to support environments like stadiums and large public venues.

Security-infused network fabric

The third principle for workplace AI involves infusing security into the network fabric. According to Dhingra, Cisco plans to build its network devices around its silicon architecture to enable more secure systems. The systems will include the following features:

• Improved trust. Creates tamper-proof systems.
• Post-quantum cryptography. Enables secure boot capabilities.
• Kernel-level compensating control. Mitigates exploits without reimaging or rebooting devices.

Infrastructure modernization is critical for AI

It's crucial for organizations to optimize their infrastructures for agentic AI. Modernization is necessary for all forms of AI, but especially for this next evolution of the technology.

Many consider AI "overhyped," Patel said, but he added that AI is being underestimated for what it can achieve, and agentic AI is going to enable AI to deliver on many of its promises.

"AI isn't just about aggregating information and coming back with answers," he said. It's about creating original insights that didn't exist in the human corpus of knowledge. When that starts to happen, we can solve problems that we couldn't have dreamed of solving before."

Deanna Darah is site editor for Informa TechTarget's SearchNetworking site. She began editing and writing at TechTarget after graduating from the University of Massachusetts Lowell in 2021.