From Intents to Actions: Agentic AI in Autonomous Networks

TL;DR

The paper presents an Agentic AI framework for autonomous, intent-driven network control in RAN, combining a lightweight, language-guided interpreter, a Bayesian-optimized interpreter–optimizer loop, and a multi-objective MORL-based controller that operates near the Pareto frontier. It introduces a dual-SLM interpreter to translate intents into an optimization template (OTM) and reason under constraints, and a PAX-BO optimizer to plan preferences across multiple services. The controller uses Distributed Envelope Q-Learning (D-EQL) to learn a single, preference-conditioned policy and scales exploration through simplex-stratified actors, with hindsight relabeling and sharded replay to improve sample efficiency. A comprehensive proof-of-concept in a 5G-compatible simulator demonstrates improved performance over traditional RL and standard LA baselines, illustrating autonomous translation of intents into near-optimal, service-aware network actions across varying conditions. The framework promises scalable, end-to-end intent management and adaptation, with potential extensions to hierarchical RAN layers and cross-domain orchestration.

Abstract

Telecommunication networks are increasingly expected to operate autonomously while supporting heterogeneous services with diverse and often conflicting intents -- that is, performance objectives, constraints, and requirements specific to each service. However, transforming high-level intents -- such as ultra-low latency, high throughput, or energy efficiency -- into concrete control actions (i.e., low-level actuator commands) remains beyond the capability of existing heuristic approaches. This work introduces an Agentic AI system for intent-driven autonomous networks, structured around three specialized agents. A supervisory interpreter agent, powered by language models, performs both lexical parsing of intents into executable optimization templates and cognitive refinement based on feedback, constraint feasibility, and evolving network conditions. An optimizer agent converts these templates into tractable optimization problems, analyzes trade-offs, and derives preferences across objectives. Lastly, a preference-driven controller agent, based on multi-objective reinforcement learning, leverages these preferences to operate near the Pareto frontier of network performance that best satisfies the original intent. Collectively, these agents enable networks to autonomously interpret, reason over, adapt to, and act upon diverse intents and network conditions in a scalable manner.

Figures (15)

• Figure 1: Agentic AI system for intent and resource management in autonomous networks.
• Figure 2: Characterization of preference-guided LA using MORL to satisfy service intents.
• Figure 3: Validation of the intent fulfillment loop between optimizer-controller for two examples.
• Figure 4: Validation of the full agentic workflow, with intent management loop and intent fulfillment loop working in unison. We compare two formulations with rigid and flexible service requirements.
• Figure 5: Dual-SLM interpreter agent. A (supervised) fine-tuned SLM generates the OTM; an intent monitor aligns telemetry; an ICL-based advisory module outputs discrete adjustment directions with rationale; and an adaptor applies bounded updates and persists them atomically. The optimizer then solves against the latest OTM snapshot, with telemetry closing the loop.

Theorems & Definitions (3)

• Remark 1: Interpretation of Linear Scalarization
• Example 1: Multi-Objective Q-Learning
• Example 2