Login
Register
This Catalyst defines a TM Forum-aligned framework that enables AI agents to safely execute real operational tasks across telecom systems, moving the industry from experimentation to scalable deployment.
Turning agentic AI into a safe, scalable operational capability
Telecom operators are under increasing pressure to move beyond conversational AI toward AI that can safely execute real operational tasks across live telecom environments. Yet many initiatives remain constrained by fragmented data, siloed APIs, inconsistent business logic, and the absence of a reusable execution and governance framework.
The challenge is not simply model performance. Operators need AI agents that can discover the right capabilities, access business context safely, orchestrate actions across BSS and OSS domains, and execute only within controlled operational boundaries, with policy control, auditability and human oversight built in.
Without that reusable approach, every AI use case becomes a separate integration effort, slowing adoption, and increasing cost. The result is a gap between AI experimentation and repeatable operational value.
The Essential framework for telecom agentic AI Catalyst, showcased at DTW Ignite in June 2026, addresses this challenge by defining a practical, TM Forum-aligned approach to operationalizing agentic AI at scale.
The Catalyst defines a governed execution framework aligned with TM Forum’s Open Digital Architecture (ODA), bringing together a tool catalog, semantic context layer, multi-agent orchestration and a structural human-in-the-loop gate before any consequential write action is performed.
At execution time, the framework gives agents standardized access to TM Forum Open APIs and AI-readable business context while preventing direct database access. A Subscriber Context Token is minted from an authoritative profile to read, enabling agents to reason on structured claims rather than raw subscriber records.
Every action is evaluated against a Cedar policy engine before any BSS system is invoked, both at session level and for each individual tool call. Each invocation also generates an immutable audit record, capturing the policy decision, agent, tool, parameters hash and outcome. A human confirmation field acts as a hard gate: if confirmation is missing, the execution agent exits before performing a write action.
The project demonstrates this architecture through representative customer care and AI-driven operations use cases, showing how operators can apply a reusable reference model within existing ODA landscapes rather than building isolated AI demos or one-off integrations.
The Catalyst builds on TM Forum’s AI-Native Blueprint member project, ODA and Open APIs, including TMF640 Service Activation, TMF677 Usage Consumption Management and TMF629 Customer Management. It also references TM Forum assets including IG1274M AI Agent, IG1355 AIOps Top Use Cases, IG1274J AI Taxonomy for Telcos, IG1430 Modern Data Architecture User Stories and Business Requirements, IG1356 Data Architecture for AI Enabled Telecom Operations and GB1087 Agentic Interaction Security in Telecommunications.
The project also proposes an ‘agentic Capabilities’ extension to ODA Canvas component definitions, enabling BSS components to declare AI-callable capabilities that can be registered into a governed tool catalog. This provides a practical path to AI-enable ODA-compliant landscapes without architectural disruption.
Based on external industry benchmarks and the team’s business-value analysis, the framework has the potential to reduce customer service costs by 15% to 20% and improve productivity across customer care and network operations by 30% to 40%. By replacing isolated integrations with a reusable platform approach, it could also shorten the deployment cycle for suitable AI use cases from months to weeks.
Trust and control are central to the design. The framework includes policy enforcement, consent control, secure partner connections, tool visibility based on authorization and tamper-evident auditability, supporting safer AI execution in environments with low tolerance for error.
Haisung Kwon, Vice President, Head of AT/DT Platform Development Division, AT/DT Center, SK Telecom, said the Catalyst is valuable because it addresses “one of the most critical gaps in telecom AI today: how to safely and reliably turn AI agents into operational actors within real telecom environments.” By grounding agentic AI in TM Forum Open APIs and ODA, he said the project provides a practical path from experimental AI pilots to scalable, governed and reusable operational AI.
For the wider industry, the Catalyst contributes a shared reference architecture and implementation pattern for governed agentic execution over TM Forum Open APIs, helping reduce fragmentation and inform future work on AI-enabled ODA components, AI governance models and operationalizing AI-Native Blueprint principles.
