Verizon pushes autonomous networks toward the customer

• Poor data and slow processing remain major obstacles to dependable autonomous networks, Praveen Atreya of Verizon said at the Ericsson OSS/BSS Summit today
• Network data masks what the customer is seeing
• But solving this challenge requires more than adding AI to existing operational processes

ERICSSON OSS/BSS SUMMIT 2026, ZURICH, SWITZERLAND – Operators should judge network performance less by aggregate metrics and more by the experience of individual customers, according to Praveen Atreya, VP of Network Systems at Verizon.

Speaking at the Ericsson OSS/BSS Summit 2026 in Zurich, Atreya said network data can show that infrastructure is functioning normally while masking problems affecting particular users — a fault of high latency in moving the data from the source of a problem to the actioning layer of the network.

“A lot of times, network data masks what the customer is seeing,” said Atreya. “How many times have you had an issue where the customer has a problem, but the network data at an aggregate level looks fantastic. … It's important for us to realize that — and realize we also have a solution for that now with AI.”

The 3-stage path to autonomous networks

Solving the challenges requires more than adding AI to existing operational processes. Instead, he described a three-stage path toward autonomy. He suggested operators begin with closed-loop automation for repetitive, domain-specific work, then introduce AI-assisted recommendations before moving to systems that can “do and notify.” 

The longer-term objective is cognitive autonomy, where agents sense network conditions, reason about them, decide on an appropriate response and act within defined guardrails, he said.

This three-stage process includes several control layers. Near-real-time reflexive loops, including Kubernetes controllers and RAN intelligent controllers, respond immediately to hardware or application problems. OSS systems form another layer for issues that escape those initial controls. An outer customer loop then brings together network, device and application information. 

Sounds logical. Well, the harder part may be making those decisions trustworthy, he said. Operators need clean, real-time telemetry and strong data governance because AI agents may take incorrect actions when presented with incomplete or poor-quality information.

Detecting a network event is only part of the job. Operators must connect that information to customer systems so they can identify affected users, communicate proactively and provide an estimated time to resolution. 

Autonomous networks as a set of nested control loops

With AI, operators have “the ability to resolve customer issues through this triage reasoning and action do more than more than we've ever had before,” he said.

He described an autonomous network as a set of nested control loops:

1. Reflexive loop: The innermost loop responds to immediate issues in near real time — for example, a system controller detecting a problem and restarting or rerouting a component.
2. Network operations loop: A broader loop detects issues that were not resolved immediately and coordinates action through operational support systems.
3. Customer-centric loop: The outer loop considers the customer’s device and applications as well as the network, helping identify issues that network telemetry alone may not reveal.

That outer loop is important for customers who experience problems but never contact support. These “silent sufferers” may not appear in complaint data, so operators need ways to identify and investigate their experience proactively, he said.

Connecting the network to the customer

Ultimately, autonomous network operations can become the connective tissue between the network, the operator and the customer, he said. When a network event affects customers, customer-facing teams and channels should have relevant information: who may be affected, what the likely impact is and what resolution timeline can reasonably be communicated.

This points to closer coordination between the OSS and BSS, which ultimately could help customer-care teams anticipate why someone is calling, acknowledge known issues and set expectations while a fix is underway.

Read more about autonomous networks on Fierce Network

HPE's self-driving network fixes 70–80% of issues on its own: Rahim

AI agents raise new control questions for telecom networks

NGMN: Agentic AI needs guardrails before it can run telco networks