Home Blog Scaling the autonomous network: introducing the Data Steward & Core Network agents

Scaling the autonomous network: introducing
the Data Steward & Core Network agents

Written by Prashant Kumar, Product Strategy Director and Naresh Rao, Product Lead at Google Cloud
Published by Google Cloud on 4 March 2026

Last year, Google Cloud unveiled the Autonomous Network Operations framework, a comprehensive blueprint designed to help Communication Service Providers (CSPs) transition from manual management to zero-touch operations. As part of that launch, we introduced a RAN AI agent as a Proof-of-Value (PoV) asset, demonstrating how generative AI could autonomously optimize radio access networks (RAN).

Today, we are announcing the evolution of our framework. Google Cloud, in partnership with Future Connections, is expanding these capabilities to move beyond isolated tasks towards agentic AI that reasons and executes across the entire telecom stack. We’re introducing two critical innovations integrated directly into the umbrella: the Autonomous Data Steward and the Core Network VoLTE Agent, which New Zealand telecom provider One NZ has deployed into its operations.

“When we launched the Autonomous Network Operations Framework, our goal was to provide a blueprint for the future of telecom. Today, we are solidifying that framework by developing and launching critical new agents under the Autonomous Network umbrella. One NZ is deploying the Core Network VoLTE Agent, which is a testament to this progress, demonstrating how we are empowering operators to move beyond isolated use cases and truly operationalize AI at scale.” – Muninder Sambi, VP & GM of Products, Google Distributed Cloud.

The industry bottleneck: why agents struggle to scale

While the industry has successfully deployed individual agents, scaling to a fully autonomous ecosystem has revealed a fundamental infrastructure challenge: data access is broken. Currently, every organization developing a new AI agent faces the same hurdles:

Data silos: data is locked in rigid schemas and scattered across domains. For example, RAN and Core in BigQuery or Probes in Cloud Storage
Semantic complexity: without semantic context, data scientists can spend months deciphering cryptic vendor-specific counters (e.g., “pmCounterNumber_487”) rather than building models
Access issues: each new agent requires a custom integration project, making rapid scaling challenging

To solve this, we must first activate the data infrastructure’s intelligence. This requires a dual-layered approach: an intelligent agent to govern the data foundation and specialized agents to act upon it.

Expanding the framework: the Autonomous Data Steward

The Autonomous Data Steward is a Gemini-powered agent that serves as the intelligent backbone of the framework. It doesn’t just store data; it dynamically provisions it on demand for other agents to use.

Built on Google Cloud, the Steward utilizes a “zero-copy” architecture. Using Dataplex Universal Catalog, it stores metadata pointers rather than duplicating massive datasets. This approach reduces storage costs by up to 70% while helping ensure agents can access real-time telemetry.

The Autonomous Data Steward further automates the DataOps lifecycle by leveraging Google’s advanced reasoning models to bridge the gap between human intent and technical execution. This enables:

Semantic understanding: when a user or agent asks for data (e.g., “cell drop rate”), Gemini interprets the natural language intent and maps it to specific telecom counters using Dataplex Universal Catalog
Dynamic ETL creation: if the dataset doesn’t exist, the Steward autonomously writes and executes Python code (via Dataproc) to extract, clean, and join the data — reducing weeks of engineering to minutes
Cross-domain unification: the agent can generate federated queries that join RAN performance, Core session failures, and Probe errors into a single view, transparently uniting isolated domains

The Core Network (VoLTE) Agent

We are moving from reactive monitoring to self-healing networks. OSS remediation agents don’t just alert humans to problems; they diagnose and solve them. By breaking down data silos, we are building a cognitive layer that turns months of engineering expertise into milliseconds of automated action.

The IMS Quality Agent (VoLTE Agent) is our specialized core network specialist. Built on the foundation of the Autonomous Data Steward, it provides excellent voice service in three ways:

Real-time monitoring: it continuously tracks critical metrics like Call Setup Success Rates, Mean Opinion Scores (MOS), and CSCF performance
Intelligent root cause analysis: powered by Gemini’s multimodal reasoning, the agent correlates complex signaling data (SIP/Diameter) with Core performance to instantly identify issues like CSCF overload
Autonomous recommendations: the agent recommends corrective actions, such as adjusting call routing or rebalancing load

When the VoLTE Agent detects a dropped call, it queries the data to instantly fetch correlated IMS signaling and probe captures — data that previously lived in disconnected systems.

One NZ is already deploying this agent to proactively manage voice quality, moving beyond manual maintenance toward a fully autonomous network.

“We’re excited to partner with Future Connections and Google Cloud, using Gemini’s advanced capabilities to deploy agentic workflows into our network. This is another meaningful progress step on our journey towards fully autonomous network operations.” – Kieran Byrne, CTO, One NZ.

Accelerating the journey to autonomous operations

These launches represent an exciting leap forward in the framework. By integrating the infrastructure intelligence of the Autonomous Data Steward agent with the domain-specific logic of the VoLTE Agent into the Autonomous Network Operations framework, we’re delivering:

For operations teams: a critical MTTR accelerator, slashing the time required to identify and resolve voice service issues. This helps teams move from detection to resolution in a fraction of the time, maintaining strong network performance
For data teams: a deployment accelerator, collapsing the time between model development and live production. This allows data teams to bypass traditional bottlenecks of data operations and realize the value of new ML deployments almost instantly

Together, we’ve open-sourced these core methodologies, allowing operators to build and customize their own agentic workflows.

Ready to get started? If you’re interested in gaining early access to the Autonomous Data Steward or the VoLTE Agent, or want to learn more about One NZ’s deployment of the agent, please reach out to your Google Account Team or Future Connections today.