Two warnings landed in the same week from opposite corners of the telecom industry, and together they say something important about where network technology is headed. One group of infrastructure engineers cautioned that so-called agentic AI – software that does not just flag a problem but goes ahead and acts on a live network on its own – needs real guardrails before it is trusted with production traffic. A separate industry association warned carriers against handing their entire AI strategy over to the handful of large cloud platforms that now dominate the conversation. Neither warning was aimed specifically at regional providers like Cytranet, but Doug Roberts, the company’s chief technology officer, says both land close to home.
“Everybody in this industry is talking about AI right now, and most of it is genuinely useful,” Roberts said. “But there is a real difference between AI that watches a network and tells a human something is wrong, and AI that reconfigures a router or reroutes traffic by itself at two in the morning. We think that difference deserves more attention than it has been getting.”
Watching versus acting
Cytranet has spent the last several years building automated monitoring into its network security and managed services, using pattern recognition to catch unusual traffic, failing equipment, or early signs of an outage before a customer ever notices. Roberts describes that layer of automation as settled technology at this point – useful, well understood, and something the company’s engineers trust. What he is more cautious about is the next step the industry is racing toward: letting AI systems take independent action on live infrastructure without a person in the loop.
“Detection is a different problem than action,” he said. “A monitoring system that’s wrong costs you a false alarm. An automated system that’s wrong and also has the authority to make changes can cost you an outage for a hospital, a school district, or a small business that can’t afford the downtime. Our government and enterprise customers are not paying us to be first in line to find that out.”
That is why, Roberts said, Cytranet’s approach has been to expand what AI is allowed to see well ahead of what it is allowed to touch. Automated systems can surface a recommendation, prioritize it, and hand it to an engineer with the context needed to act quickly – but for now, a person still makes the call on anything that changes how traffic actually moves. It is a slower path than letting software run unsupervised, and Roberts is comfortable with that trade-off.
Why independence matters as much as intelligence
The second warning, about carriers outsourcing their AI future to a small number of hyperscale platforms, touches a different nerve for a company built around local, personalized support. Cytranet has spent over a decade serving businesses, nonprofits, and government agencies with its own engineering team rather than a call center script, and Roberts argues that the value of that model only grows as more of the industry’s intelligence gets concentrated in a handful of distant platforms.
“If every regional provider just plugs into the same third-party AI platform and lets it make the decisions, you lose the thing that made regional providers useful in the first place, which is that we actually know our own networks and our own customers,” Roberts said. “We use outside tools where they make sense, but the judgment about what’s right for a school district in a rural county or a manufacturer running a night shift has to stay close to the ground. That’s not something we’re willing to hand off entirely.”
Roberts is careful to note this is not a rejection of AI, and he pushes back on the idea that caution and adoption are opposites. Cytranet continues to invest in automation across its fiber network, managed IT, and network security offerings, and he expects the pace of that investment to pick up, not slow down, over the next year. The guardrails, he said, are what make faster adoption possible rather than what holds it back.
Building the guardrails now, not later
Concretely, Roberts said that means a few standing rules inside Cytranet’s own operations: automated systems get read access to the network long before they get write access, every AI-assisted recommendation is logged and reviewable, and any tool that touches customer-facing infrastructure goes through the same change-management discipline as a human engineer would follow. None of that is glamorous, he acknowledged, but he thinks it is the difference between adopting AI responsibly and adopting it quickly.
“I’d rather our customers never notice the AI at all,” Roberts said. “The best outcome is that their internet stays up, their calls go through, and their data is safe, and none of that requires them to think about how it happened. If we’ve done our job, the technology is invisible and the guardrails are the reason it stays that way.”
As agentic AI moves from an industry talking point to something carriers are actually deploying, Roberts said the companies that earn long-term trust will be the ones that can explain, in plain language, what their automated systems are allowed to do and what still requires a human decision. For a company whose customers include government agencies and businesses that cannot afford unplanned downtime, he considers that an easy trade to make.
