The promise of agentic networking is straightforward: give artificial intelligence more context about the network, let it interpret events across systems and help operations teams respond faster.
The reality, according to Zayo’s SVP of Network Connectivity Max Clauson, is more cautious.
While AI agents may increasingly support network teams with incident investigation, reporting and event correlation, Clauson said the technology is not yet at the point where enterprises should allow it to independently decide on and execute consequential network changes.
That distinction sits at the centre of the emerging debate around agentic networking. As providers look to connect AI agents with network information and management tools, they must show that automation can improve service outcomes without handing over control of critical infrastructure.
“Agentic networking” may have become a popular label across the sector, but Clauson said the market should not treat it as a single, settled category.
Three Stages of AI-Driven Networking
Zayo sees AI-driven networking developing across three broad areas, Clauson said.
The first is AI chat and knowledge tools. These can help users query information, navigate documentation and access reporting more easily. The second is AI-supported network operations, in which agents assist existing teams with tasks such as assessing performance data and investigating issues. The third, and most advanced, is agentic networking supported through Model Context Protocol (MCP) servers, allowing AI tools to integrate more directly with networking systems.
“We really see AI-driven networking or agentic networking in three different buckets,” Clauson said.
“The first is AI chat, the second is sort of AI-supported network support, and the third, sort of most advanced element, is really getting into MCP-supported agentic networking.”
He expects the first category to become standard over time, while AI assistance for network teams will become more widespread. But the third category remains at the frontier of the market.
“[It] needs to have time to digest and move forward.”
Human Approval Remains Central
The limits of autonomy are particularly significant during a network incident.
Outages and performance issues often involve incomplete telemetry, conflicting alerts and pressure to restore service quickly.
An agent might be able to correlate events across applications, cloud environments and network infrastructure far faster than a person. But what happens when the recommendation is wrong, or when an authorised action makes the problem worse?
Clauson said agents are not yet broadly operating without oversight.
“I don’t think we’re yet to a point where agents are working truly autonomously, where they’re making recommendations and then executing on those recommendations as well,” he said.
Instead, he said, agents can assess conditions, make recommendations on provisioning or identify patterns during an incident. The human network team retains responsibility for deciding whether to act.
“The agents will propose a change, but then would need confirmation before moving forward with taking any action on the network itself,” Clauson said.
For enterprises, that model may be a more realistic starting point for adopting AI in service operations. Rather than attempting to automate every step of incident response or network change, organisations can use AI to speed up analysis while preserving approval gates for actions that could affect availability, security or compliance.
Control Depends on Customer Appetite
Zayo is positioning that degree of control as a customer choice.
Clauson said some organisations do not want AI involved in their network operations at all, either because they believe the technology is too immature or because they are not confident that governance controls are sufficient. In those cases, the underlying networking services remain available without an AI layer.
At the other end of the spectrum, some customers are attracted to products that offer AI-supported operations because they see an opportunity to give smaller teams more capability or make faster assessments during live service issues.
