As mobile networks evolve toward 5G and 6G, the RAN (Radio Access Network) is becoming too complex for manual configuration and static optimization. Enter RAN Intelligent Controllers (RICs) - the cornerstone of the O-RAN Alliance architecture.
RICs host applications that bring intelligence and automation into the RAN. These applications are of two kinds: xApps and rApps. At first glance, they might sound similar, but they play ery different roles in the automation journey.
An xApp lives inside the near-real-time RIC. Its job is to make fast, localized decisions in the RAN, typically on the order of milliseconds to one second. This means an xApp can directly influence what happens on the radio interface, deciding which user gets scheduled on which resources, triggering handovers, or mitigating interference between neighboring cells. Because of this tight control loop, xApps interact with the distributed RAN nodes (like O-DUs and O-CUs) through the E2 interface. In simple terms, you can think of an xApp as the reflex system of the RAN, acting instantly to keep the network running smoothly.
An rApp, in contrast, runs in the non-real-time RIC, which is part of the Service Management and Orchestration (SMO) framework. rApps operate on longer timescales - seconds to hours, sometimes even days. They are not designed for reflexive actions but for strategic oversight and optimization. rApps consume data from the network, identify patterns, forecast demand, and refine configuration policies. They also handle the intent-based management paradigm, which is increasingly central to network automation.
Where Agentic AI Fits. Agentic AI refers to Large Language Model (LLM)-driven software agents that can read plain-language intents, plan multi-step actions, call network tools/APIs, and verify outcomes. In O-RAN, think of agents as an “intent-to-action” layer that orchestrates - not replaces - rApps and xApps.
With intent handling, rApps allow operators to express what they want the network to achieve, without dictating how it should be done. For example, an operator might declare an intent such as “ensure ultra-reliable low latency for factory automation devices in this region” or “guarantee minimum throughput for premium slice customers during peak hours.” The rApp translates this high-level intent into concrete policies and objectives, which are then passed to the near-RT RIC over the A1 interface. The xApps, in turn, enforce these policies in real time by adjusting scheduling, resource allocation, or mobility decisions. This closes the loop between high-level goals and real-time actions, bridging the gap between strategic planning and operational execution.
From Intent to Action. An AI agent can accept free-text goals (e.g., “prioritize URLLC for factory devices in Region A from 08:00–18:00”), translate them into a concrete plan (draft policy, simulate impact, schedule the change), and push the policy via A1 to the near-RT RIC. It can then monitor live KPIs, compare results to the stated intent, and iterate. The xApps handle the sub-second enforcement on the radio side; the agent keeps the loop aligned with the operator’s goals.
Where Agents Run. For planning, assurance, and orchestration, agents typically live in the SMO / non-RT RIC alongside rApps. For very tight loops, lightweight Small Language Models (SLMs) may assist inside the near-RT RIC (e.g., fast rule selection or anomaly triage), as long as latency and determinism requirements are met. In all cases, agents complement rApps/xApps and use existing A1/E2 guardrails and policies.
The interplay between rApps and xApps is what makes RAN automation possible. xApps ensure that the network reacts instantly to local changes, while rApps provide the intelligence to align those local actions with global objectives and long-term efficiency. Agentic AI sits between intents and policies, turning high-level goals into executable actions and coordinating which rApps/xApps to use. Together, they create a closed-loop system: xApps collect and act on live metrics, rApps analyze and refine intents and policies, and agents plan, validate, and iterate on those policies as the network learns and adapts.
This division of labor has profound implications for operators. It reduces the need for manual tuning, cuts operational costs, and accelerates innovation by allowing specialized vendors to deliver modular apps. It also enables advanced services like SLA-driven slicing and AI-powered optimization, where business-level requirements expressed as intents can be automatically translated into technical actions. Agentic AI strengthens this translation by accepting natural-language intents, drafting and simulating policies, enforcing guardrails (approvals, rollback, audits), and when latency permits, using lightweight SLMs for near-real-time assistance, complementing rApps/xApps.
Practical note. Agentic AI in telecom is still early, but it’s a strong accelerator for intent capture, policy drafting, troubleshooting, and cross-domain automation - best introduced with guardrails and human-in-the-loop approvals.
Learn more by reading our xApp and rApp Communication Guide.