Agentic AI Is Reshaping Telcos from the Inside Out

Telecom operators are moving into a new phase of AI adoption in which the story is no longer about adding more automation to existing processes, but about changing the operating logic of the network and customer-facing systems themselves. According to Eoin Coughlin, Global CTO and Industry Lead for Telco at IBM Software, for ICTBusiness Media - ICTbusiness.info, the industry is shifting toward an agentic model in which multiple specialized AI agents collaborate, decide, and orchestrate the next best action across the network. That, he argues, creates room to tackle problems that have long remained beyond the reach of conventional OSS and BSS tools.

Coughlin says traditional autonomous networks were “more deterministic”, built around predefined thresholds and fixed insights, while agents now make a much higher level of autonomy possible. “The paradigm is shifting now to an agentic model,” he says, describing an orchestration layer that can decide, based on the data it sees, which agents to invoke next, whether for data collection, interpretation, or model-driven decision-making. In practical terms, that means operators can move away from rigid step-by-step automations and toward systems that can analyze, create, and act more fluidly.

A major part of that shift is the recognition that general-purpose LLMs are not enough on their own. “LLMs are not good at managing time and understanding network data,” Coughlin notes, which is why IBM is using time-series models tailored to the realities of network operations. Once those models are combined with manuals, trouble tickets, historical evidence, and operational documentation, agents can do more than surface anomalies. They can start to reason about what is actually happening. That is where, in his view, OSS begins to change: not by replacing everything operators already have, but by adding a new layer alongside existing systems. “You don’t want to rip all of that out and start again straight away,” he says. The goal is to address what current toolsets still cannot see.

That matters particularly in the case of so-called silent network issues, from configuration anomalies to problems that never appear on standard dashboards. “We’re trying to solve problems which are unsolvable today,” Coughlin says. The difference comes from the combination of context, computing, and agents that can both generate and analyze rather than simply execute fixed instructions.

At the infrastructure level, IBM ties that transformation to the longer move toward containerization, microservices, and hybrid multicloud environments. Coughlin argues that it became clear years ago that virtual machines alone would not be enough in a world where applications run on-premises, in private or air-gapped environments, and across multiple clouds at once. That is why he points to Red Hat and OpenShift as foundational building blocks, giving operators a consistent layer for infrastructure management, application deployment, security, and automation across different environments.

The BSS side may be where the commercial impact becomes most visible. Coughlin points to invoice analysis as an example, where AI can identify customers at risk of bill shock before frustration turns into churn. “About 12 per cent of people who get bill shock actually leave an operator within two months,” he says. If AI can flag unusual charges early, explain unfamiliar bill items, and let customers interact with an agent before the case reaches a human adviser, operators can cut service costs while improving the overall customer experience.

Yet Coughlin arguably sees the biggest opportunity beyond the traditional connectivity role in helping telecom operators become providers of sovereign AI infrastructure for enterprises, government bodies, and smaller businesses. As concerns around sovereignty intensify and latency become critical for agentic AI, computing needs to move closer to where data is used, at the edge and inside telecom data centers. “The reason why you want it at the edge is that you want it close to where the enterprise is using the information,” he says. In his view, operators already possess several of the ingredients required for that role: infrastructure, data centers, and an understanding of tough regulatory environments. “An operator can build a multi-tenancy stack” for customers that want access to AI without building their own complex platforms, all within guardrails for governance, trust, and security. That is why agentic AI in telecom is emerging not simply as another technology wave, but as a potential new operating and market model for the sector.