The days when a dropped call or slow data speed sent teams of engineers racing to diagnose and repair the problem may soon belong to history.

At Mobile World Congress in Barcelona this week, Google Cloud demonstrated the next major step toward truly autonomous telecom networks. Its latest agentic AI systems go beyond alerting operators to issues—they actively sense problems, reason about the best fixes, and carry out corrective actions with zero human touch.

This work expands on the Autonomous Network Operations framework Google introduced in 2025. The long-term ambition is clear: reach Level 4 to 5 autonomy, where the network itself can "identify, diagnose, and fix its own problems without human intervention."

A Living Model of the Network

The foundation of this approach is the network digital twin. Google has transformed what used to be a static diagram into something far more powerful:

We’ve evolved the digital twin from a static map into a temporal graph powered by Cloud Spanner Graph.

This dynamic, time-aware model continuously reflects the network’s real physical and logical state—performance metrics, fault conditions, traffic flows, everything. Agents can look backward in time (querying how the network appeared five hours or five days earlier) to instantly understand root causes. They can also run forward-looking simulations: what happens if a fiber cut occurs here, or if we deploy this upgrade there?

A new data steward agent ensures the twin stays perfectly synchronized with reality by automating governance and validation tasks.

Agents That Don’t Just Watch—They Act

Google also showcased specialized autonomous network agents now in trials with several operators. These handle specific domains such as voice core networks and operational support systems (OSS). When trouble appears, they don’t wait for instructions. Examples of zero-touch actions already demonstrated include:

• Automatically rerouting traffic around a failed link using the best remaining path
• Adjusting parameters in real time to restore call quality when degradation is detected

Trials with operators like One NZ have shown these agents successfully resolving issues that would previously have required manual intervention.

To help the wider industry adopt this technology faster, Google has open-sourced its telco data pipeline and core data models on GitHub. The release removes much of the manual mapping pain, letting operators implement standardized ontologies and build accurate digital twins more quickly.

Real Operators Are Already Moving

Deutsche Telekom is one of the frontrunners. In partnership with Google Cloud, it has developed MINDR—a multi-agent system powered by Gemini models. MINDR watches across radio access, transport, and core domains, predicts emerging risks, and autonomously applies fixes before customers are affected. It builds on the earlier RAN Guardian Agent, which has been optimizing radio networks during major events since 2025, with broader deployment across Germany expected later this year.

Other collaborators include Vodafone, MasOrange and NetAI (working together on GraphML-based AI operations), and Nokia, whose Network as Code platform now integrates with Google’s agentic stack. That integration lets engineers describe desired outcomes in natural language—“prioritize bandwidth for emergency services in this region for the next two hours”—and have the network reconfigure itself accordingly.

High Stakes, Important Questions

Telecom networks are among the most intricate and mission-critical infrastructures we rely on. They run 24/7, support millions of users simultaneously, and outages can cost operators millions of dollars per hour while damaging customer trust.

For decades the industry has talked about self-healing networks. At MWC 2026 we’re seeing concrete pieces of that vision in production trials: multi-agent systems, accurate digital twins, and genuine autonomous remediation.

Yet the shift also prompts reflection. When an AI agent silently corrects a problem deep in the network core, who carries ultimate responsibility if the fix introduces an unforeseen side effect? And as routine operations move toward full automation, how will the role of network engineers evolve—from hands-on troubleshooting to higher-level strategy, design, and ethical oversight?

Google frames the transition as augmentation rather than replacement: dramatically better reliability, sharply lower operating costs, and the ability to deliver smarter, more responsive connectivity services.

After years of promises, the agentic telco is no longer just a concept. The demonstrations in Barcelona show it is entering the real world—one autonomous fix at a time.

Primary sources: Google Cloud blogs (“Autonomous networks at MWC 2026” and “Google Cloud and the rise of the agentic telco”), Deutsche Telekom MWC 2026 announcements, Nokia statements, SiliconANGLE, Fierce Network, and TelecomTV coverage.