Why SMO Must Evolve for the Agentic AI Era
- Telecom Infra Project
- 21 minutes ago
- 4 min read
The next generation of network automation will require more than adding AI models to existing operational workflows.
As networks become more distributed, cloud-native, and interconnected across RAN, transport, core, and edge environments, operators will need systems that can interpret intent, coordinate decisions across domains, and adapt as network conditions change.
A new white paper from the Telecom Infra Project’s OpenRAN Project Group examines how the O-RAN Service Management and Orchestration framework could evolve to support that future.
SMO Evolution & Agentic AI, developed by the RAN Intelligence & Automation Subgroup, explores how Agentic AI could help transform SMO into an intent-driven, cognitive control layer for increasingly complex networks.
The limits of today’s SMO environment
The O-RAN SMO provides an important foundation for open and intelligent RAN operations. Its scope includes network function management, O-Cloud resource orchestration, service assurance, analytics, policy management, and AI/ML workflows. However, the current SMO ecosystem remains uneven.
AI capabilities are often introduced through individual applications or isolated use cases. Cross-domain orchestration remains limited, and many automation workflows are still localized rather than coordinated across the full network.
The white paper identifies several areas that will require further development, including:
Cross-domain intelligence spanning RAN, core, transport, cloud, and edge
More mature policy-driven and intent-based automation
Consistent approaches to AI and machine learning lifecycle management
Support for federated learning and secure data collaboration
Stronger interoperability across SMO services and interfaces
Unified workflows for assurance, optimization, fault handling, and orchestration
These gaps become more significant as the industry looks toward 6G. Future networks will need to manage far greater operational complexity while supporting increasingly demanding service-level, performance, energy, and resilience requirements.
Incremental automation alone is unlikely to be enough.
Moving beyond predictive AI
Many current AI-enabled network applications are designed to perform a defined task. They may forecast traffic, detect an anomaly, optimize a parameter, or recommend an operational change. Agentic AI introduces a different operating model.
An AI agent can interpret a goal, break it into a sequence of tasks, gather the required information, use available tools, coordinate with other agents, and determine what action should happen next.
Within an evolved SMO, specialized agents could operate across individual domains while working toward a shared network or business objective. A planning agent could translate an operator’s intent into an execution plan, while RAN, transport, core, and cloud agents coordinate the required actions within their own environments.
This does not mean allowing AI systems to make unrestricted changes to live networks. The white paper proposes a governance model that includes policy enforcement, security controls, auditability, human escalation, and a Policy Guardian Agent responsible for validating proposed actions before they are executed.
The objective is governed autonomy: greater operational speed and coordination without removing accountability or control.
Where Agentic AI could make a difference
The paper examines two use cases that show why this approach matters.
Energy optimization across the network
Reducing network energy consumption requires operators to balance several competing considerations.
Infrastructure may support different low-power modes depending on the equipment vendor. Energy prices and carbon intensity can vary over time. Any change must also account for coverage, capacity, mobility, subscriber experience, and service-level commitments.
Optimizing one domain in isolation can simply move the cost or performance impact elsewhere.
An agentic system could reason across RAN infrastructure, compute resources, transport, traffic demand, energy pricing, and operational policies. It could coordinate a series of actions while continuously assessing their impact on network performance.
This creates the potential for energy optimization at a system level rather than through a collection of disconnected rules.
Cross-domain fault resolution
Network faults rarely respect organizational or technology boundaries.
A customer experience issue may originate in the RAN, transport network, O-Cloud, core, or a combination of several domains. A single incident can also generate hundreds of related alarms, making it difficult for operators to identify the underlying cause.
An agentic approach could correlate alarms, suppress downstream symptoms, assess urgency, and run diagnostic and mitigation activities in parallel.
For example, the system could redirect traffic to protect service levels while continuing to investigate the fault. It could maintain a record of every action taken, why it was taken, and what network state changed. Once the issue is resolved, that record could support an ordered rollback of temporary mitigations.
When human intervention is needed, the agent could provide a technician with a targeted diagnostic plan and incorporate the field results into the ongoing investigation.
This creates a shared operational workflow between autonomous systems and human experts rather than treating the two as separate processes.
A decision point for 6G architecture
The industry is now entering an important period for 6G architecture and standards development.
The white paper argues that Agentic AI should be considered during the 6G study phase, while the underlying principles of future network management and orchestration are still being defined.
Waiting until specifications and architectures are largely complete would risk treating Agentic AI as an additional application layer. That could leave the industry attempting to retrofit goal-driven, multi-agent automation into an SMO framework that was not designed to support it.
Considering Agentic AI at the architectural level creates an opportunity to address intelligence, governance, data, policy, security, and orchestration together.
Turning architectural ideas into deployable systems
Standards organizations will continue to define the architectural and technical foundations for future networks. The next challenge will be translating those specifications into interoperable systems that operators can deploy and manage.
TIP’s role is to help connect these areas.
Through the OpenRAN Project Group, TIP can bring operators, suppliers, cloud providers, and system integrators together to develop practical use cases, reference designs, implementation requirements, test plans, and validation frameworks.
That work will be essential to determine how Agentic AI behaves across multi-vendor environments, how autonomous actions are governed, and how new capabilities can be introduced alongside existing 5G and Open RAN deployments.
The goal is not autonomy for its own sake. It is a network operating model capable of responding to complexity with greater coordination, efficiency, resilience, and trust.
Read the SMO Evolution & Agentic AI white paper to explore the proposed vision, use cases, architectural considerations, and TIP’s role in supporting its development.