Matilda Bailey is a distinguished expert in next-generation networking, specializing in the architecture required to support the rapid onset of the Agentic AI era. As organizations pivot toward artificial intelligence, the underlying IP networks often become the forgotten constraint, yet Matilda argues that modernization is the only path to true monetization. This conversation explores the necessary shift from static, legacy frameworks to dynamic systems characterized by real-time telemetry, segment routing, and flexible algorithms. We delve into how these advancements allow for precise path control and automated policy enforcement, ensuring that high-stakes industries like finance and healthcare can scale their AI ambitions without compromising on security or performance.
The traditional “busy hour” traffic model has defined networking for decades, but you suggest it’s no longer relevant in an AI-driven world. How does the shift to an “always-on” environment fundamentally change how we build and manage these systems?
We are seeing a complete dissolution of the predictable peaks and valleys that once allowed operators to catch their breath and plan for maintenance. In this new era, AI agents are hitting the network around the clock, operating on a heartbeat of continuous demand that never truly ebbs. These agents aren’t just moving data; they are making critical decisions in microseconds, which places an immense, non-stop pressure on the infrastructure. To survive this, we have to move away from networks designed for the relatively slow pace of human-initiated voice and video. Instead, we must build a dynamic ecosystem capable of adapting resources instantly across multi-cloud environments, ensuring that the connectivity is as fast as the intelligence it carries.
When we talk about the speed of AI, the human element can often become a bottleneck. How does implementing real-time telemetry change the day-to-day experience for network operators who are used to manual troubleshooting?
For too long, network operators have been flying blind, forced to rely on static reports that are essentially ancient history by the time they hit someone’s desk. It is incredibly frustrating to deal with a performance dip when you are looking at data that is minutes or hours old, especially when AI workloads require instant responses. By integrating real-time telemetry, we shift from a reactive, firefighting stance to a proactive model of automated intervention. This visibility allows the network to sense a shift in traffic patterns as it happens, enabling the system to self-correct before a human even realizes there’s a problem. It replaces the frantic manual search for errors with a streamlined, intelligent oversight that matches the sheer velocity of the AI agents it serves.
Modernizing a bloated IP architecture is a daunting task for any large organization. How do segment routing and EVPN allow for a more graceful evolution without requiring a complete “rip and replace” of existing investments?
The beauty of segment routing is that it acts as a bridge, leveraging the hardware and investments an organization already has in place while stripping away the unnecessary complexity of legacy protocols. In the old world, a network architect might have needed weeks of careful planning and manual configuration to adjust for new demands, which is an eternity in today’s market. With a foundation built on segment routing and EVPN, those same changes can be executed within seconds, providing the precise path control needed for modern AI connectivity. It turns a rigid, cumbersome structure into a fluid environment where convergence isn’t just a goal, but a daily reality. This transition is less about a catastrophic overhaul and more about a strategic evolution that keeps an organization relevant as its connectivity needs scale.
FlexAlgo seems to be a game-changer for traffic engineering, specifically regarding how it handles diverse SLAs. Can you explain how this capability allows a network to “think” about the best path for a specific workload?
FlexAlgo is the intelligence layer that allows us to move past a one-size-fits-all approach to data routing. We can define specific performance objectives, such as minimizing latency for an AI decision or maximizing bandwidth for a massive data transfer. It also allows us to bake in resiliency and meet strict data sovereignty requirements automatically, ensuring that sensitive information stays within specific geographical or policy boundaries. Unlike the old RSVP-TE method, which was a nightmare of manually engineered tunnels and constant state management, FlexAlgo lets the operator set the constraints and then lets the network do the heavy lifting. It results in a much more elegant, automated way to ensure that every AI agent gets exactly the performance it requires without being constrained by static rules.
You’ve worked with enterprises in the healthcare and finance sectors where the stakes for data integrity and speed are incredibly high. How are they successfully balancing high-performance AI demands with the need for rigorous security like MACsec?
These sectors are the ultimate proving grounds for next-gen networking because they cannot afford a millisecond of compromise in either speed or security. By incorporating MACsec directly into their modernized architecture, they are able to encrypt traffic at the hardware level without sacrificing the low-latency performance that AI workloads crave. We see these organizations managing a complex mix of AI-driven services and traditional workloads, all while adhering to ironclad sovereignty and SLA requirements. Some choose to manage their own IP networks over leased optical lines to maintain total control, while others prefer the agility of a fully managed service model. The end result for both is a resilient, secure foundation that prevents connectivity bottlenecks and ensures that their digital transformation doesn’t outpace their ability to protect their data.
What is your forecast for the future of AI-integrated networking?
I believe the gap between leaders and laggards will be determined entirely by the agility of their IP networks over the next few years. We are moving toward a future where the network is no longer a passive pipe, but an active participant in the AI ecosystem that automatically enforces business policies and performance goals. Those who successfully monetize the AI moment will be the ones who stopped looking at connectivity as a utility and started treating it as a strategic asset. If you stand still and rely on the rigid protocols of the past, you risk being completely overrun by competitors who have embraced this evolution. The network will become the ultimate differentiator, acting as the nervous system for every intelligent agent and automated service in the enterprise.
