AI System Automatically Generates Network Simulations for Autonomous Management

Network operators seeking true autonomous management systems face a critical bottleneck: existing digital twins require engineers to hardcode analysis logic for every scenario. A new approach sidesteps this constraint by using language models to generate network simulators on demand.

The Challenge of Autonomous Networks#

Modern telecom and data center networks aspire to self-managing autonomy—systems that can evaluate business rules and policies in real time without human intervention at each decision point. The TM Forum Level 4 autonomy framework envisions this maturity level, but achieving it requires evaluating Network Service Intents (NSIs) across constantly changing conditions. Building a new simulation model for each scenario remains impractical.

Network Digital Twins are the natural solution: executable behavioral models that allow operators to run what-if analyses before committing to changes. The constraint is that traditional NDTs embed their decision logic in fixed code, making them inflexible when requirements shift.

NDT Factory: AI-Driven Synthesis#

Researchers have introduced the NDT Factory, a multi-agent software platform that generates executable network digital twins automatically from semantic descriptions using large language models. Rather than hand-crafting simulation code, operators provide a model of their network and policies in a declarative format; the system then orchestrates parallel synthesis and generates a complete, compilable, and testable NDT.

The team validated the approach using a Call Admission Control (CAC) case study—a classic telecom problem where a system must decide whether to accept or reject new calls based on current network load and constraints. The NDT Factory synthesized a full CAC simulator and achieved impressive results:

• 100% compilation and test pass rates across multiple independent runs
• 99.3% agreement with a reference implementation when simulating 300 test cases
• 90% admission rate and correct attribution of all rejections, showing that the generated logic faithfully captured the intended policy

Why It Matters#

This approach addresses a real pain point in autonomous network management: the gap between policy intent and execution. Rather than manually implementing analysis logic, operators can specify what they want to happen and let AI generate verifiable code. The deterministic, testable output means the generated simulators can be audited and trusted—critical for production networks where errors carry real costs.