Table of Contents
Verizon CTO Yago Tenorio details an AI strategy based on the knowledge that “makes us unique’
Verizon is publicly targeting Level 4 automation in critical parts of its core network, and said its closed-loop automation platforms executed more than 70 million autonomous configuration changes in 2025. But CTO Yago Tenorio drew a useful distinction between automation at scale and autonomy. The former is about doing known things faster. The latter requires a system that can reason through situations it was not explicitly programmed to handle, identify root causes and decide what to do next.
“At the end of the day, we want AI to be able to do what our people do, but faster so that we can get more stuff done,” Tenorio explained in a discussion with RCR Tech. The important part of that statement is not faster. It is what our people do.
“You need to train AI with skills that summarize our experience, expertise in running a network. That’s what we think makes us unique,” he said.
The strategic asset, in other words, is not access to the latest frontier model. Those models are advancing rapidly and becoming broadly available. “The intelligence of the system is almost out of question,” Tenorio said. The differentiator is whether an operator can provide sufficiently granular data, relevant context and codified operating expertise.
That idea becomes concrete in how Verizon is rethinking customer experience data. Networks have traditionally viewed experience from the inside out through huge numbers of KPIs. But the absence of a network event does not mean the absence of a customer problem. A customer in a basement speakeasy with no usable signal may not register at all. “By definition the network doesn’t even know you exist, so it’s not captured in any KPIs,” Tenorio said.
Verizon is working to complement network-side data with device telemetry and associate performance indicators with precise locations. Tenorio described using a smartphone’s altimeter data to infer where in a building a customer is experiencing a problem. The significance is the creation of a physically grounded picture of customer experience rich enough for AI systems to reason over.
Verizon is now prototyping what Tenorio called “a training camp of agents,” with specialized agents examining potential root causes across transport, core and radio domains. The system remains human-supervised. Tenorio said Verizon hopes to put the radio component into production by the end of 2026, with transport and core to follow. In prototype cycles involving real customer-impacting network issues, he said diagnosis that previously took engineers five to eight hours has been reduced to around 90 seconds. Verizon’s public description of the architecture similarly centers on permanent, specialized agents working across network data to diagnose and remediate problems.
This is where high-level cognitive automation begins to shade into autonomy. The agents are being trained on operator skills, given domain-specific scope and asked to reason collectively across a problem.
A network graph becomes critical context. Mapping network elements and their relationships allows an agent to understand that an action on one element may affect several others. As Tenorio put it, the graph provides the context for understanding, “if I do X in element one, then A, B, C are going to get impacted.”
Verizon is building its agent platform internally, with a control plane that assigns identities, scopes and permissions to agents. Some may be allowed to create other agents, but not without limits. This is a practical model of bounded agency wherein an agent can act, but within declared authority and guardrails.
That internal AI strategy also extends to human workers. Tenorio said Verizon’s rollout of Claude Code to 33,000 people is, in effect, turning a much larger portion of the organization into software developers. Much of the resulting software is not intended to become customer-facing applications. Instead, employees are building abstraction layers over years of accumulated internal systems and operational complexity.
In this sense, Verizon is attacking the autonomy problem from two directions. AI agents are being trained to reason and act, while humans are using AI to simplify the environment those agents will operate in.
Tenorio is notably pragmatic about the industry’s autonomy taxonomy. Verizon uses TM Forum’s methodology to benchmark domains and processes, but he does not want reaching a particular level to become the objective. That is probably the right tension. Maturity frameworks establish a baseline and common vocabulary. Customers care whether problems are found faster, fixed earlier and prevented from recurring.
Verizon’s near-term job is to make the network aware enough of itself, its customers and the consequences of its own actions that AI can move from recommending to doing. The longer-term bet is that the same combination of sensing, context, reasoning and edge execution will reshape what the network itself is built to connect.
