Over 90 service providers worldwide have already launched or are preparing for the commercial rollout of 5G Standalone (SA) networks, according to the Ericsson Mobility Report (November 2025). This technological transition has drastically increased the complexity of infrastructure management. The telecommunications industry must move beyond simple text-based AI chatbots toward autonomous AI agents. However, autonomous networks cannot function solely on AI without human oversight—they require deep integration into the IT landscape to perform routine operations.
In practice, deploying such agents faces a fundamental obstacle: data silos and rigid vendor lock-in. An AI agent cannot correctly perform a task if it is isolated from the real-time state of the network or the subscriber's current balance. Traditional monolithic BSS/OSS systems are not designed for microsecond requests from agents, and the lack of a unified data orchestration bus turns every integration into a complex and slow custom project.
Why AI chatbots in telecom fail without deep BSS/OSS integration
Most modern AI implementations in telecom are limited to customer care. Assistants can recognize user intent, but they lack integration with back-office systems to execute actions, such as dynamically changing service configurations or activating additional network slices.
For autonomous operation, AI agents require real-time access to Business Support Systems (BSS) and Operations Support Systems (OSS). Without this, "swivel-chair automation" occurs, where AI only generates a request, while an operator performs the final data entry into billing or provisioning systems. True orchestration requires agents to independently initiate transactions via standardized APIs.
TM Forum ODA architecture as a countermeasure to vendor lock-in
Monolithic BSS/OSS systems were built for static services, featuring closed databases and proprietary interfaces. Attempting to connect an AI agent to them usually leads to an unmanageable web of point-to-point integrations. To solve this, TM Forum developed the Open Digital Architecture (ODA) concept. It involves moving away from monoliths toward a componentized, decoupled architecture with an API-first design.
This approach aligns perfectly with the service-based architecture (SBA) of 5G core networks, based on cloud-native principles (per 3GPP standards). It is important to understand that ODA is not an "out-of-the-box" solution for immediate implementation. It is a multi-year strategic modernization roadmap that requires the consistent exposure of billing and network monitoring functions through standardized Open APIs.
Requirements for the billing core: millisecond response and DooxSwitch integration
When an AI agent initiates a tariff plan change for a B2B client, integration latency must be measured in milliseconds. Traditional batch processing is unacceptable in such scenarios. The agent must verify limits, calculate costs, and authorize the transaction in real time without manual intervention.
In the telecom landscape managing voice and IoT/M2M traffic, an example of a system capable of providing the necessary speed is the carrier-grade VoIP platform DooxSwitch. Thanks to its built-in real-time billing core and LCR routing mechanisms, this platform allows for transaction processing with minimal latency. Using ODA-compliant APIs as a bridge between the client AI interface and DooxSwitch billing enables the secure automation of commercial operations.
Data orchestration: the role of the UnityBase low-code platform in eliminating data silos
Even with fast billing, data often remains scattered: CRM stores interaction history, while OSS stores network topology. An AI agent needs a single logical orchestration layer that hides the complexity of the physical infrastructure.
To build such a layer, the low-code/model-driven platform UnityBase can be used (a joint development by companies of the Intecracy Group alliance, where InBase is a key, but not the only, developer). Using a unified domain metadata model, the platform automatically generates REST API, providing AI agents with a standardized interface for accessing various DBMS. Thanks to its asynchronous non-blocking HTTP server, UnityBase ensures high integration performance, eliminating the need for thousands of lines of custom code. For high-load systems or increased security requirements, Enterprise (EE) or Defence (DE) commercial editions are applied.
Cybersecurity: protecting new attack surfaces per ENISA requirements
Implementing AI agents does not eliminate the need for robust security protocols; on the contrary, agents create an additional attack surface. According to the ENISA Threat Landscape 2025 report, digital infrastructure and services account for approximately 27.7% of all recorded data breaches. Compromising an AI agent in telecom could lead to unauthorized network configuration changes or access to personal data.
The ODA architecture requires the implementation of the Zero Trust concept at the API gateway level. UnityBase (in EE/DE editions) provides these mechanisms at the platform core level, supporting detailed audit trails, Role-Based Access Control (RBAC), Access Control Lists (ACL), and Row-Level Security (RLS). This ensures that an AI agent has access only to the minimum data required to perform a specific transaction.
| Comparison criteria | Point-to-point custom integrations (spaghetti code) | Componentized decoupled ODA architecture |
|---|---|---|
| Billing access speed | High latency (seconds), batch processing | Real-time (milliseconds) via standardized API |
| Vendor dependency | Full lock-in; any change requires vendor development | Shared API contracts, easy component replacement |
| Data security level | Shared database access (high breach risk) | Role-based and attribute-based access (RLS/ACL) at the API gateway level |
FAQ
How does TM Forum ODA help integrate AI agents into billing without risking network stability?
ODA replaces monolithic architecture with decoupled components using an API-first design. AI agents interact with billing exclusively through standardized and secure Open APIs, minimizing the risks of direct database interference.
What are the technical requirements for the DooxSwitch billing system API for working with autonomous agents?
The billing system API must support millisecond response times and real-time rating integration. DooxSwitch provides this capability through an optimized real-time billing core and modern interfaces.
How can subscriber personal data leaks be avoided when granting AI agents access to OSS/BSS?
It is necessary to implement a Zero Trust architecture. Using platform mechanisms such as Row-Level Security (RLS) and Access Control Lists (ACL) in commercial editions of UnityBase allows for strictly limiting the visibility of data for an AI agent based on its current task.