How does this agentic AI approach differ from the automation telcos already use?

Current telecom automation typically executes predefined scripts and runbooks for known problems, corresponding to Level 2-3 autonomy. The agentic approach aims for Level 4-5 autonomy, where agents can independently understand operator intent, research unencountered problems, propose and evaluate multiple solutions using digital twins, and execute coordinated actions across network domains under strict policy governance, effectively moving from a reactive to a proactive and investigative operational model.

How Telcos Build Autonomous Networks with Agentic AI

From Automation to Autonomy: Telcos Build the Brains for Self-Driving Networks

Telecommunications operators are beginning to shift from isolated automation scripts to integrated platforms for genuine network autonomy. While most telcos currently operate at Level 2-3 on the TM Forum’s autonomy scale, the push towards Level 4-5 requires a new architecture where AI agents can sense, reason, and act cohesively. This transition depends on a unified platform built on shared domain models, policy controls, and digital twins, a domain where technology providers like NVIDIA are introducing a foundational stack to enable this next generation of network operations.

The proposed architecture organizes AI agents around a 'problem-solution loop' to handle different operational challenges. This involves 'on-demand agents' for simple tasks, 'long-running agents' for persistent problem management, and 'deep research agents' to explore novel issues without predefined solutions. This system is underpinned by a comprehensive technology stack designed for the telecom domain, enabling agents to safely reason and act. Key components of this platform include:

Data & Models: Technologies like NVIDIA NeMo Data Designer for synthetic data, Nemotron for reasoning models, and NV-Tesseract for time-series analysis create the telecom-aware foundation.
Agent Orchestration: The NVIDIA Agent Toolkit provides the framework to build and manage agent workflows, connecting them to shared tools and evaluation systems.
Secure Execution: A secure runtime, such as NVIDIA OpenShell, isolates agent processes in sandboxes, while blueprints like NemoClaw manage agent deployment and governance according to strict policies.
Deep Research: Specialized frameworks like the NVIDIA AI-Q blueprint allow a team of agents to fan out, gather evidence from disparate systems, and propose ranked solutions for complex problems.

The impact of this agentic approach extends beyond streamlining current operations to enabling autonomous innovation. In a practical example, a 'deep research agent' can analyze a network issue like SR-MPLS tunnel degradation, propose multiple remediation plans with trade-offs, and pass them to a 'long-running agent' for policy-governed execution and monitoring. More significantly, this agentic framework is being applied to research and development. The NVIDIA AI Telco Engineer project demonstrated that an agentic system could autonomously discover novel wireless network algorithms for link adaptation that outperformed established industry standards, showing that these platforms can expand a network's capabilities, not just maintain them.

The strategic shift for telcos is not simply adopting more AI tools, but architecting a unified autonomy platform where persistent, policy-governed agents can compound their knowledge, moving from executing known solutions to autonomously discovering and validating new ones.

>> Verify Original Transmission at NVIDIA