Agent AI will force rethinking at the edge of the network

Artificial intelligence enters a new phase with agent AI: autonomous systems that perceive, decide, act and learn without constant human supervision, operating independently in all distributed environments while interacting with other agents in real time.

This transition from centralized AI models to distributed, autonomous agents requires a rethinking of the architecture of the wide area network infrastructure. Previous AI patterns, such as centralized training clusters, cloud-based inference and hub-and-spoke data flow, are insufficient for agent systems that must operate at the network edge with speed, autonomy and resilience.

In these environments, a WAN is no longer just a way to connect branch sites to central data centers. It becomes the critical fabric that enables edge agents to synchronize data, share information and coordinate actions, making WAN performance, availability, and flexibility critical to agent AI performance.

Edge-centric intelligence

Consider a self-driving car navigation system, a smart manufacturing facility or a retail environment where AI agents manage inventory, pricing and customer information simultaneously.

While WAN connectivity enables agents to synchronize across locations, edge locations often face unpredictable connectivity. In such cases, agents can perform automatic remediation during WAN degradation by ensuring real-time path selection for critical systems such as point of sale, asset synchronization and IoT devices. These are sequential decisions that need to be made in milliseconds based on local conditions, often when connections to centralized systems are interrupted or blocked.

Unlike traditional AI models that operate on data in controlled environments, agent systems exist in a virtual world where delays are measured in milliseconds, and decisions have immediate consequences. Sending data hundreds of kilometers away to a cloud data center for processing is not possible. An agent must process information, evaluate options, and act locally – right where the action is taking place.

In addition, agent AI systems often operate in environments with multiple agents that communicate across distributed environments. A smart city deployment may include thousands of agents that control traffic flow, energy distribution and public safety simultaneously. These agents need to share data even when the network connection goes down.

Integrated computing

To work, agent AI needs computing resources integrated with data sources and decisions, which means sending highly efficient processing to thousands of distributed locations in industries including retail, manufacturing, healthcare, and transportation.

These computing resources must deal with various workloads, agents making rapid predictions on distributed data, making fine adjustments to the spatial model based on environmental feedback and communicating with peer agents. In retail, this may translate into supporting smart shelves, computer vision systems, digital signage, loss prevention analytics and customer flow optimization directly at each store location.

A powerful edge computer alone cannot deliver the full potential of agent AI. Without equally complex communication, autonomous agents remain isolated, unable to collaborate with peers, synchronize information or maintain intelligence gathered across distributed locations.

A high-performance network integrated directly into the edge computing infrastructure enables agent-to-agent communication with low latency and high bandwidth, instead of routing all communication through remote connection points. This construction method, where the network and the computer are built together, is important for real-time collaboration.

Security is equally important. These systems require the cryptographic identity of every agent, encrypted communications, hardware-based trust roots and zero-trust architectures designed in both layers from the ground up, which ensure the integrity of autonomous decisions affecting physical systems and the safety of people in critical infrastructure such as healthcare and transportation.

Meeting at the edges

Organizations cannot simply expand cloud infrastructure to set up environments and expect agent systems to succeed. The autonomous, distributed, real-time nature of these systems requires an infrastructure designed to support local intelligence, agent interactions and secure operations in thousands of different locations.

Equally important is end-to-end visibility to the edge. As organizations deploy distributed AI agents across large, disparate environments, continuous visibility into WAN performance, network health and application performance at each edge location becomes critical. This allows teams to proactively identify issues, improve performance, and ensure reliable service delivery. Blind spots undermine autonomy and AI needs to be robust.

Organizations making infrastructure choices today will decide whether they lead this change or spend years retooling. This requires rethinking both edge deployment and WAN evolution to support distributed intelligence at scale. Computing convergence and communication at the edge is a critical foundation for the next generation of autonomous, intelligent systems.

Lee Peterson is vice president of secure WAN product management at Cisco Systems Inc.

Photo: theCUBE Research/DALL-E