Architecting autonomy: Network infrastructure for the agentic era

• Achieving network autonomy: Verizon is targeting Level 4 cognitive automation to build an intelligent core network that reasons and self-heals.
• Scale and efficiency: Closed-loop automation platforms executed over 70 million network configuration changes autonomously in 2025.
• Frontier model integration: Engineers embed advanced language models like Anthropic’s Claude to manage traffic and optimize performance.

For most of the telecommunications industry’s history, the network was treated as a static asset. Engineers calculated its raw capacity, deployed fixed hardware infrastructure, and reacted to issues after the fact through flashing terminal alarms and reactive troubleshooting dashboards.

The cloud started to slowly change the static model, but as we are catapulting into a macro-era defined by generative, agentic, and physical AI, yesterday’s static connectivity models are now quickly becoming obsolete. What we are building today is categorically different: AI is becoming the operational control system of the network itself.

At Verizon, we are redefining our future by building our networks purposefully and intentionally to serve as the preferred connectivity foundation for the world’s most responsive AI solutions. We are moving away from traditional hardware constraints to turn our infrastructure into a software-defined, self-organizing reasoning engine that optimizes itself in real time intentionally. We are not just adopting tools; we are engineering an AI-native enterprise.

To understand where we are today in the journey toward true network autonomy, consider an automotive analogy. Right now, standard telecommunication systems sit at a conceptual “cruise control”—relying heavily on static scripts and conditional, human-led automation. Verizon’s target, however, is to achieve high-level cognitive automation. We are pushing hard toward Level 4 autonomy in critical segments of our core network.

There is a sharp distinction between running automated scripts on a rigid schedule and true, decentralized autonomy. Autonomy means building an intelligent system that can reason about an unexpected, complex situation it was never explicitly programmed for, isolate the hidden root cause, and execute an architectural solution completely out of the human loop.

We are delivering the first phases of this autonomous vision at an industrial scale today:

• 70+ million actions: In 2025 alone, our closed-loop automation platforms executed over 70 million network configuration changes autonomously.
• Efficiency realized: Implemented by hand, these granular changes would have required thousands of technicians working around the clock.
• Capital reinvestment: This intelligent engine reclaims thousands of operational labor hours and drives millions in direct capital savings, allowing our workforce to focus on high-value engineering judgment rather than digital chores.

To reach our target of Level 4 autonomy, we have doubled down on GenAI. We are now embedding frontier language models like Anthropic’s Claude into our internal operations, effectively democratizing software development across our entire engineering workforce.

Our teams are utilizing natural language to handle complex modeling, proactively manage surging traffic, preempt service issues, and optimize the performance of every connection. Engineers no longer need to write traditional syntax or lines of custom code; they simply define the high-level architectural outcome they want to achieve. This paradigm shift has allowed us to build an organization-wide repository of shared expertise, human nuance, and methodology that sits directly on top of our unified data layer.

To realize our autonomous strategy, we have moved beyond centralized rules to a live hierarchy of independent, permanent AI agents running directly on our infrastructure. We are currently piloting an Autonomous Network program with intelligent agents running 24/7. These agents spot complex data patterns and take swift action on anomalies—such as changing configurations, resetting network elements, or opening high-priority tickets when human intervention is needed for resolution - based on real-time data.

A master agent can immediately detect a localized anomaly, take control, and instantly spin up a network of specialized sub-agents focusing on the exact root cause, pinpointing it down to the precise network domain and specific element. Now, what used to take engineering teams hours to manually diagnose and correct is identified and resolved in less than two minutes—long before the end-user ever experiences a dropped call or a lag in their data connection. 

An autonomous system, however, is only as good as the live data feeding it. Traditional Network KPIs tend to skew toward highly favorable conditions and don’t always fully capture the true, localized end-user experience. Our new AI-driven models close that critical telemetry gap. By leveraging aggregated, anonymized network performance signals—including floor-by-floor indoor environments—we build an accurate, hyper-granular picture of service quality.

This rich physical telemetry, fused seamlessly with existing network data, feeds a continuous optimization loop. It detects and resolves coverage and capacity issues proactively, often before a customer is impacted. The result is near-real-time visibility into in-building performance environments, enabling automated alerting and network tuning at a granularity we’ve never possessed before.

By feeding an advanced reasoning engine with uncompromised, granular physical network signals, we are building an autonomous architecture that does not just manage raw connectivity—it completely redefines what a network can be.

If you enjoy experimenting with technology, you’ve probably experienced some transformative personal journey with AI. My favorite one was building an AI running coach to make me a better, faster, less injury-prone runner. To me, it proved that raw intelligence is no longer the digital bottleneck; the true differentiator in the next wave of technology is frictionless access to big pools of relevant data and physical edge execution. What I am seeing on an individual scale with running performance metrics is exactly what Verizon is delivering on a massive, industrial scale for the global technology ecosystem.

By the time 6G arrives around 2029 or 2030, this intelligent infrastructure will achieve full physical, contextual awareness. This will allow future wearable devices to implicitly map and interact with the real-world environment surrounding you in real time. These context-aware AI agents will require an Uplink Revolution, to support constant streaming of massive volumes of high-definition video and sensor telemetry upward to edge inference models. 

To get there, open standards like O-RAN and deep ecosystem interoperability are non-negotiable architectural foundations. We have already broken past traditional vendor lock-in, successfully running multi-vendor AI orchestration applications simultaneously across live production platforms. We are moving fast, shifting permanently away from the reactive paradigms of the past to build a software-defined, self-healing digital fabric. At Verizon, we aren’t just adapting to the future of technology—we are engineering the intelligent foundation that makes it possible.