Key facts

• Agentic AI requires networks that adapt in real time as AI agents request data, trigger actions, and collaborate across distributed, multi-cloud environments.
• Traditional 'busy hour' traffic patterns are being replaced by always-on demand from AI workloads.
• Legacy IP networks are too rigid and complex; segment routing and EVPN provide convergence and precise path control.
• FlexAlgo allows networks to automatically compute optimal paths based on latency, bandwidth, resiliency, or data sovereignty requirements.
• Real-time telemetry is essential for automated intervention and moving away from reactive manual troubleshooting.
• Organizations that modernize IP networks can monetize AI services, while those that hesitate risk losing ground.

Agentic AI changes the traffic model

Agentic AI represents a shift from single AI requests to autonomous systems that plan and execute multi-step tasks with minimal human intervention. These AI agents do not simply wait for a user prompt and return an answer. They interact with applications, APIs, databases, and other agents. They make decisions, request data, and trigger actions across distributed, multi-cloud environments. That means the network is no longer carrying occasional bursts of traffic from humans browsing the web or streaming video. Instead, it is carrying constant, machine-generated traffic from agents that operate around the clock.

The old traffic model of predictable 'busy hour' peaks is fading. In the past, network architects could design capacity around known periods of high demand, such as a morning email surge or an afternoon video conference spike. AI agents, however, are active at all hours. They gather data, run inference, update models, and execute workflows continuously. This creates an always-on traffic profile with continuous demand. A network built for human-centric traffic patterns cannot keep up with the performance, speed, and agility required by AI workloads.

The challenge is not just bandwidth. It is also about dynamic behavior. AI agents often need to communicate with each other in microseconds. A delay of even a few milliseconds can impact the quality of an AI-driven decision or the responsiveness of an autonomous workflow. Traditional networks were designed to deliver voice, video, and general internet traffic. They were not designed for the unpredictable, latency-sensitive, and highly distributed communication patterns of AI agents.

Real-time telemetry and autonomous operations

One of the most important requirements for an AI-ready network is real-time telemetry. Operators need to understand traffic patterns as they happen, not hours later. Without real-time information, network teams are forced to rely on reactive manual troubleshooting. They might look at static reports, but those reports are often outdated by the time they are reviewed. In an environment where AI agents are continuously requesting data and taking actions, stale information is nearly useless.

Real-time telemetry gives operators visibility into sudden changes in traffic flow, latency spikes, packet loss, and congestion. It also enables automated intervention. Rather than waiting for a human to notice a problem and open a ticket, the network can automatically reroute traffic, apply quality-of-service policies, or alert the appropriate system. This is essential for supporting AI workloads that require consistent performance and high availability.

The need for automation goes beyond fault management. AI agents are dynamic, and their connectivity needs change frequently. A network that relies on manual configuration and static policies will struggle to keep up. With real-time telemetry and automation, the network can respond to changing conditions in seconds. That is a major departure from the past, when network architects often had weeks to plan and implement changes. In the agentic AI era, network conditions must change within seconds to meet the requirements of AI agents.

Segment routing and EVPN as the foundation

To meet these new demands, organizations need to evolve away from bloated, rigid, and complex IP architectures. Legacy IP networks and traditional protocols served enterprises well during earlier eras of VPN and internet connectivity. But they are too rigid and too complex for dynamic AI workloads. Modernization often starts with segment routing and EVPN, which together provide a solid foundation for convergence and precise path control.

Segment routing is particularly valuable because it leverages existing network investments. It does not require a complete overhaul of the underlying infrastructure. Instead, it creates an evolutionary path to flexibility. Segment routing allows operators to define explicit paths for traffic without the complexity of older protocols. It makes it possible to steer traffic dynamically as AI agents' connectivity needs change.

EVPN, or Ethernet Virtual Private Network, complements segment routing by providing scalable, flexible multi-tenancy and network virtualization. It enables the separation of different workloads and services, making it easier to support a mix of AI and traditional applications. Together, segment routing and EVPN enable dynamic traffic routing and help operators simplify their networks. This simplification is essential for AI because complex, manual processes cannot scale at the speed of autonomous agents.

The old way of engineering traffic was labor intensive. Network engineers would manually configure paths, manage state on every router, and carefully coordinate changes. With modern segment routing and EVPN, the network itself can compute paths based on intent. Operators define the desired outcome, and the network handles the mechanics. This reduction in complexity is critical for organizations that want to deploy AI services quickly and reliably.

FlexAlgo: matching traffic to performance requirements

Another key capability is FlexAlgo, short for 'flexible algorithm.' FlexAlgo lets the network calculate optimal paths for different traffic types based on defined constraints and performance objectives. For example, one class of traffic might be optimized for low latency, another for available bandwidth, another for resiliency, and another to satisfy data sovereignty requirements. This is important because not all AI workloads have the same needs.

A real-time AI agent that is interacting with a customer might need a path with very low latency. A background data processing job might care more about available bandwidth than latency. A compliance-sensitive workload might need traffic to stay within a particular geographic boundary or follow a specific path for regulatory reasons. FlexAlgo allows operators to define these performance objectives and constraints, then lets the network automatically compute and maintain the appropriate paths.

In many ways, FlexAlgo delivers the traffic-engineering benefits that operators once sought with RSVP-TE, but without the massive complexity. RSVP-TE relied on manually engineered tunnels and extensive state management. Every tunnel had to be set up, monitored, and torn down, creating a significant operational burden. FlexAlgo simplifies this by using the network itself to calculate paths based on algorithms. Operators can define different algorithms for different traffic classes, and the network automatically routes traffic accordingly.

As networks increasingly support different service-level agreements for different AI agents and workloads, FlexAlgo ensures that traffic is matched to performance requirements rather than constrained by static, one-size-fits-all rules. This dynamic matching is essential for organizations that need to offer differentiated services to internal teams or external customers. It also helps prevent connectivity bottlenecks by making optimal use of available network resources.

Security: MACsec and beyond

Modernizing the network for agentic AI is not just about performance and routing. It is also about security. AI agents often operate across distributed, multi-cloud environments, which means their traffic traverses a variety of network segments. Protecting that traffic from interception and tampering is critical, especially in sectors such as healthcare and finance.

MACsec is one security capability that can be incorporated into the network architecture. MACsec provides encryption at the data link layer, protecting traffic as it moves between network devices. It is designed to be high throughput and low latency, making it suitable for AI workloads that require both speed and security. By adding MACsec, organizations can ensure that AI agent communications remain confidential and integrity-protected without creating a significant performance burden.

Security is not only a technical requirement; it is also a business requirement. In regulated industries, organizations must demonstrate that they can protect sensitive data and meet compliance obligations. A modern network architecture that includes MACsec and other security features can support these requirements while still enabling the agility needed for AI innovation.

Real-world adoption in healthcare and finance

Large enterprises in critical sectors such as healthcare and finance are already beginning to incorporate these capabilities into their network architectures as part of broader digital transformation initiatives. They need to support a mix of AI and traditional workloads while ensuring that traffic adheres to strict policy, sovereignty, and service-level requirements.

Some of these organizations choose to deploy and manage their own IP networks over leased optical services from providers. This approach gives them direct control over their network architecture and allows them to customize policies, security, and performance for their specific workloads. Others prefer to consume the same capabilities through a fully managed network service. This model creates new opportunities for service providers to deliver differentiated, value-added services to enterprise customers.

The result is a network that automatically enforces business policies and performance objectives. It prevents connectivity bottlenecks and maintains service assurance as AI adoption and digital transformation efforts continue to scale. Instead of constantly firefighting network issues, operations teams can focus on optimizing AI outcomes and improving business services.

The competitive urgency of network modernization

AI creates both an opportunity and a challenge for service providers and large enterprises. The opportunity is the ability to monetize the next wave of AI services. Organizations with modern, flexible networks can offer new products and experiences that rely on real-time AI agents. They can provide low-latency, reliable connectivity to internal teams and external customers. They can differentiate themselves in the market by delivering AI capabilities that competitors cannot match.

The challenge is that standing still carries real risk. The pace of AI adoption is accelerating, and competitors that embrace network evolution will be able to respond faster, scale more easily, and offer richer experiences. Organizations that delay network modernization may find themselves unable to support the AI applications their customers and employees demand. They may be forced into expensive, urgent upgrades under pressure, or they may lose relevance entirely.

The network is no longer just a pipe that carries data. In the agentic AI era, it is a strategic platform that enables intelligent decision-making, automated workflows, and new business models. Those who invest in modern IP networking, real-time telemetry, segment routing, EVPN, FlexAlgo, and strong security will be well positioned to lead. Those who wait risk being left behind.

Source:Network World News