Why The Agentic Economy Will Break Today’s Enterprise Networks

Ivo Ivanov is the CEO of DE-CIX.

getty

​At 9:20 a.m. on a Tuesday, a supply chain system notices a disruption emerging thousands of miles away. Before a human manager has even logged on, software agents have already rerouted shipments, renegotiated warehouse capacity, adjusted inventory forecasts and updated customer delivery estimates in real time. In another part of the world, a cybersecurity platform isolates suspicious network activity in milliseconds, while AI-driven trading systems respond to geopolitical events faster than any analyst could process a headline. This is the new reality—a global economy that moves faster than even the most dedicated and experienced humans.​

The leap from generative AI to agentic AI is at least as significant as the arrival of AI itself. These systems don’t just generate content, crunch numbers or respond to prompts—they can operate autonomously, make decisions, coordinate with other systems and continuously adapt within defined parameters. Gartner analysts predict that by 2028, "at least 15% of day-to-day work decisions will be made autonomously through agentic AI."

The problem is that most enterprise network infrastructure was designed around human workflows and human reaction times. When machines need to communicate with each other at near instantaneous speeds, the network becomes the make-or-break factor. No matter how good the model is or how intelligent the AI, deployments will fall short of expectations if they can’t exchange data quickly and securely.​

It's Not Just About Automation

The clue’s in the name—we appreciate artificial intelligence for its intelligence. But its true value is speed. Research published in Scientific American estimates that human cognition operates at roughly 10 bits per second, or 100 milliseconds per bit, which is roughly the speed at which we talk. An AI agent, on the other hand, can carry out advanced reasoning in a matter of milliseconds, depending on the model.

Basically, there’s an enormous gap between how quickly information enters the human brain and how quickly we can consciously act on it. AI agents don’t have that limitation. They can query multiple systems simultaneously, analyze huge volumes of data in parallel and execute decisions in milliseconds without fatigue, distraction or delay. In industries like high-frequency trading, microseconds already carry financial value. Agentic systems push that principle into almost every corner of enterprise operations.​

That has profound implications for the infrastructure underneath modern business. For years, enterprise networks were largely designed around predictable patterns of human behavior, such as office hours, user logins, scheduled data transfers, video meetings and relatively linear and predictable data traffic. But AI agents operate continuously, communicating constantly with other systems while generating far more traffic between clouds, platforms, APIs and edge environments.

An autonomous logistics platform may need to exchange live telemetry between factories, vehicles, weather systems, inventory databases and financial software simultaneously. A healthcare AI assistant may pull data from imaging systems, patient records, cloud-hosted models and wearable devices in real time while coordinating with other AI systems behind the scenes. And by that point, latency matters. In fact, it becomes the defining characteristic of business success.​

Why Fast Data Means Better Resilience

What defines strong AI infrastructure? The quality of the GPUs? The size of hyperscalers? The sheer power of compute? Nvidia became one of the world’s most valuable companies because investors understood that AI would require enormous processing capability. But as agentic systems mature, a blind spot has started to emerge in how data is moved. AI agents exchange data constantly, often while carrying out time-sensitive tasks, and no matter how good they are at performing those tasks, they can’t perform if the network itself isn’t performing.​

This is forcing enterprises to rethink how their infrastructure is interconnected. Public internet routing was never designed for machine-speed collaboration between autonomous systems operating across multiple environments simultaneously. Every additional network hop introduces latency, unpredictability and potential points of failure. A manufacturing platform coordinating robotics across facilities, or a financial institution running autonomous fraud detection, can't afford bottlenecks caused by congested or inefficient data pathways.

As a result, businesses are increasingly prioritizing direct interconnection, software-defined networking and private data exchange environments that allow them to control how critical traffic moves between clouds, platforms and users. In the process, networking is becoming less about binary connectivity and more about strategic control over how data moves. Although having the “best” or “biggest” AI model seems desirable, it counts for little if the data it depends on can’t be exchanged with speed, efficiency and precision.​

Where The Risk Sits In Agentic AI Deployment

For all the buzz surrounding agentic AI, there’s some alarming evidence that organizations are underestimating what it will truly take to deploy it successfully at scale. In June 2025, Gartner analysts predicted that "over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls." At the same time, enterprises are continuing to pour money into it. Research from 3Gem found that businesses deploying agentic AI are already reporting measurable gains, including improved productivity, faster decision making and lower operational costs.

The economic incentives are real, but a lot of organizations are stuck trying to layer autonomous systems onto legacy infrastructure. That’s going to create a form of technological debt that many businesses will struggle to settle.​

Legacy architectures, fragmented cloud strategies, inconsistent security policies and poorly integrated systems all become far more problematic when decision making starts happening at “mach speed” (machine speed). Businesses often talk about AI transformation as though deploying the model is the hard part. But the bigger challenge may be redesigning the underlying infrastructure that allows autonomous systems to collaborate reliably in real time.​

U.S. political advisor James Carville once quipped, “It’s the economy, stupid!” alluding to the fact that all of America’s challenges revolve around fixing the economy. Soon, business consultants might well be saying, “It’s the network, stupid!” for very similar reasons.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?