Operationalizing intent-based autonomy at Level 4+

Commercial context

To reach Level 4+ autonomy, CSPs must build operational systems that act with genuine intelligence. A step up from automation, they need to be able to interpret business goals, simulate network responses, make decisions, and execute change with minimal human input. To get there, they need integrated intent-driven architectures, tested in live conditions and aligned to open standards.

The Autonomy Accelerated: Intent to impact - Phase II Catalyst – led by Nokia, with contributions from tech partners Qvantel and Red Hat, as well as CSP champions Telstra, Ooredoo, Telin, e&, VNPT, and Mobily – sets out to demonstrate this architecture in a live environment. It focuses on turning high-level intent into meaningful network action, using AI, digital twins, and TM Forum assets to build closed-loop systems that can manage, assure, and optimize services with minimal human intervention.

The solution

The team explored two high-impact use cases. The first is AI driven sales process acceleration for service delivery, which validates service feasibility using intent probing. The second is service assurance which includes intent assurance using closed loop automation, AIOps and Agentic AI. In each case, the solution goes beyond isolated automation to show how intent-based, AI-powered systems can work together to support scalable, trustworthy Level 4+ operations.

The solution is underpinned by an intelligent closed-loop architecture that spans the full autonomy cycle, from intent discovery to execution. It combines intent-based APIs, digital twin simulations, and agentic AI, all orchestrated using TM Forum Open APIs and ODA-aligned design.

In the first use case, real-time service feasibility, Qvantel’s Flex BSS provides an AI assisted sales process, and as part of it, it uses TMF633 for requesting available intents, and in the further steps of the sales process, initiates an intent probe using TMF921. Nokia’s Digital Operations Center receives the request and -assesses the feasibility of the proposed service using the combination of service digital twin, network digital twin and knowledge graphs. The simulation includes live and predicted network states, planned events, and contextual data. Based on this, it returns the feedback together with best possible options based on expected network conditions. This process reduces failed activations, boosts sales confidence by ensuring services can be delivered before orders are confirmed as well as creating an opportunity for further upsell.

The second use case focuses on autonomous service assurance. Traditional approaches rely on post-failure diagnostics and manual intervention. Here, the Catalyst enables continuous monitoring using AI-powered observability. Intelligent agents detect anomalies, perform root cause analysis using LLMs and retrieval-augmented generation (RAG), and propose resolution actions. These actions are validated through digital twin simulation then safely deployed via orchestration with an added layer of safety provided by human approval. By combining prediction, validation, and execution, the system prevents incidents before they affect customers.

The whole process demonstrates full intent-based lifecycle management. It uses a combination of AIOps and Agentic AI to manage each phase: fulfilment, assurance, and optimisation. Specialist AI agents collaborate using knowledge graphs, digital twins, and orchestration systems. The architecture elevates human roles from operators to overseers, providing a reliable blueprint for hands-off, scalable autonomy.

Across both use cases, the Catalyst integrates key TM Forum assets. TMF633 is used for intent service discovery. TMF921 enables intent probing and validation. TMF645 supports service qualification using forecasted data. APIs such as TMF639, TMF638, and TMF060 enable real-time resource, service, and knowledge management. The architecture also draws from TM Forum’s ODA, intent modelling guides, and genAI specifications.

Application and wider value

The Catalyst provides clear operational benefits. It accelerates the sales process, reduces failed service activations, shortens resolution times, and increases confidence in fully autonomous assurance. AI agents now manage service operations with minimal manual input, moving beyond automation towards true autonomy. Closed-loop decisions are based on predicted outcomes and digitally validated actions, removing guesswork and reducing risk.

Early results point to a 90% improvement in time-to-market for new services. Mean time to detect issues dropped by more than 90%, and mean time to repair by over 80%. AI-driven execution has cut operational workload and helped reduce OPEX through fewer manual errors and faster, more accurate decisions.

This project could mark a turning point, transforming the long-held vision of self-driving networks into something measurable, modular, and operationally viable. Hakim Agung Ramadhan, project lead and Head of Product Innovation & Development for PT at Telin, explains. "We have accelerated Level 4+ autonomy with agentic AI, digital twins, and intent-based architecture to reduce costs by at least 30% through automation." The Catalyst helps move the industry from intent capture and into end-to-end lifecycle management, backed by AI and orchestration. Overall, the team estimates the combined solution could result in USD 800 million in annual financial benefits.

For the wider industry, the Catalyst provides a viable framework for Level 4+ autonomy. It shows how intent-driven operations can evolve from isolated pilots to deployments. The approach reduces complexity by aligning business logic, network simulation, and AI-driven execution within a standards-based architecture.

The path to autonomy is complex, but this Catalyst shows that it is achievable through a modular, incremental approach. With TM Forum standards as the foundation, and agentic systems as the engine, CSPs now have a working model for delivering customer-aware, self-operating networks at scale.