Telecom operators exploring autonomous networks need accurate and continuously updated information about network resources before AI agents can make operational decisions, according to Ericsson.

In its recent blog, Ericsson said fragmented or outdated service and resource inventories can limit closed-loop automation. AI agents rely on this information to understand which network resources exist, how they are configured, and how services depend on them.

The accuracy of that data becomes more important as operators move toward higher levels of network autonomy. TM Forum defines Level 4 autonomous networks as using intent-driven and predictive decision-making, with AI models and continuous learning supporting closed-loop management of services and customer experience.

In a closed loop, network software can identify a condition, determine an action, execute the change, and assess the result. Ericsson said inaccurate inventory can affect both the decisions made by AI agents and the network actions triggered through closed-loop automation.

Ericsson said incorrect inventory data can result in systems provisioning services on unavailable resources, identifying the wrong cause of network faults, or planning capacity based on outdated information. It also cited cases where stale topology information could lead to incorrect service-level calculations or cause a digital twin to represent a network state that no longer exists.

Fragmented data complicates network automation

Network inventory records physical, logical, and virtual resources across the network. This can include sites, nodes, links, network functions, spectrum, network slices, configurations, service states, and the relationships between those resources.

Ericsson said inventory systems supporting autonomous operations need to remain synchronised with changes taking place across the network. Data can be collected from network management systems, controllers, orchestration platforms, and assurance systems before being reconciled into a common operational model.

The fragmentation problem extends beyond equipment from different suppliers. Nokia and Databricks said in June that telecom networks typically rely on hundreds of siloed operational and business support systems, with individual systems using their own data architectures.

Their proof of concept tested a common data platform designed to ingest network information in real time and make it available to AI agents making automated decisions across network domains.

Topology accuracy is another factor in automated operations. Ericsson said outdated topology can cause AI agents to misidentify affected services, available redundant capacity, or dependencies on equipment undergoing maintenance.

Multi-vendor environments add another requirement. Radio access, transport, IP, optical, cloud, and core network systems can expose operational information using different models, requiring inventory platforms to normalise the data before it is used across network domains.

Ericsson said different vendors and network domains can expose operational information through different schemas and semantics, requiring AI systems to reconcile those differences before making decisions across domains.

Standards bodies are also addressing this interoperability problem. ETSI has been developing standardised interfaces and data models for network management, automation, and information exchange across multi-vendor environments.

Nokia has similarly identified multi-vendor support, network discovery, reconciliation, and topology management as requirements for inventory systems used in autonomous networks.

Autonomous networks need context, not just inventory

A unified inventory does not contain all the information required for autonomous decision-making. Ericsson said inventory provides the operational view of what services and resources exist, their configuration, state, and relationships, while autonomous decisions also depend on information such as intent, policies, constraints, service and customer context, historical evidence, and domain rules.

Ericsson describes a knowledge plane as the layer that combines those sources using shared semantics and ontologies. Under that model, inventory maintains the network’s operational map, while the knowledge plane connects it with other information that AI agents require when making decisions.

TM Forum has identified a similar requirement. A report published in July said reaching Level 4 autonomy requires operators to move beyond managing individual data sources and processes toward using contextual information that AI systems can apply across network domains.

TM Forum describes the knowledge plane as a layer that turns fragmented network data into information that supports intent-driven operations and cross-domain autonomy.

Digital twins place similar demands on network data. They provide virtual representations that can be used to model network behaviour and assess changes before those changes reach the production environment.

ETSI is also working on network digital-twin specifications covering standardised architectures, data models, interfaces, testing, and closed-loop verification across multi-vendor environments.

Ericsson said a digital twin built on outdated inventory can represent a network state that no longer exists, which can affect its use in planning and automated network operations.

Shared data does not remove coordination risks

Even when autonomous systems operate with current network information, multiple agents can still propose actions that conflict with one another. TM Forum examined this issue in a July Catalyst involving agents, applications, orchestrators, and service functions operating over shared infrastructure.

One example involved an energy-saving action on one mobile cell shifting traffic onto a neighbouring cell that was supporting a critical service. Another involved a new business network slice competing with an existing priority commitment for available network resources.

TM Forum’s project uses a conflict-management layer to evaluate proposed actions against live service commitments before they reach the network. The system considers intents, policies, resource constraints, and actor permissions to determine whether an action could affect services, slices, cells, customers, or service-level commitments.

If an action conflicts with those constraints, the system can return the reason to the AI agent. The agent can then adjust factors such as timing, scope, or resource selection before submitting the action again.

Ericsson’s autonomous-network approach also connects inventory with intent management, closed-loop automation, network data, and knowledge models. Its September blog said AI agents require an accurate operational view of resources and services before they can take part in autonomous network decisions.

(Photo by Denny Müller)

See also: Ericsson CTO: Preparing cellular infrastructure for AI agent traffic

Want to discover how IoT is transforming telecoms and connectivity? Join the IoT Tech Expo in Amsterdam, California, and London. Explore how innovations in 5G, edge computing, and IoT are shaping the future of networks and services. The event is part of TechEx and co-located with other leading technology conferences, click here for more information.

Telecoms Tech News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.