The role of telecoms networks, and the components of them, have been shifting for some time now. The role of the TM Forum has, too. Where 20 years ago the focus would have been firmly on service delivery, billing systems and more, now the organisation is taking a lead on network automation and autonomy.
In the run-up to 2026’s DTW event in Copenhagen, TelcoForge spoke to CTO George Glass about the sizeable opportunities, but also the very real problems it carries with it.
Good Intentions
The first thing Glass wants to establish is the distinction between automation and autonomy in the network. Glass paraphrases Henry Ford:
“You’re not taking your existing network processes and trying to make them go faster. that’s training faster horses. You’re changing the way you actually operate and manage your networks. Hence, we’re designing or building tractors.”
According to him, there are two ways autonomous networks change things. The first is from the telco’s network operations perspective.
“Autonomy is where AI plays a key role in delivering the outcome that the users of your network want or the experience that you are achieving to get with them… the zero trouble, the zero wait, the zero outages from the customer’s perspective,” he said.
“And that can only be achieved if you can actually manage your network autonomously. You use AI to make the decisions to correct the underlying problems with your network that are giving perceived issues to your users.”
The other element is more nuanced, but is essentially about changing the function of the network in delivering services. Glass describes today’s networks as establishing a connection between one point and another, then making sure that there are systems and failsafes in place to keep that working as well as possible.
“With autonomous networks I want to talk about the outcome that we want, the intent, not the technology that’s used to deliver it. And I want an intelligent network that they’ll then select the best technology based on the intent that we have.
“And that can only happen whenever you’ve got concepts like intent in your network; you’ve got closed-loop controls; you’ve got AI configurators and orchestrators that are actually making those decisions about how you actually connect A and B over your network,” the TM Forum added.
A-I-A-I-O
While “AI” as a term is about as nebulous as “the network”, Glass does have very definite thoughts about the role of different types of AI in delivering autonomy. We’ve all heard stories about rogue AI agents, LLMs and chatbots – enough to make anyone serious about critical national infrastructure weep. Despite that, the hype in the industry is… problematic, to say the least.
“At the moment, it’s almost like ‘The answer’s an agent! What was your question?’ If I’m trying to make a decision, that’s got two outcomes, A or B, agentic is not needed. I don’t need to pay for tokens to make that. A rules-based decision or a data-driven decision is perfectly valid.”
Glass does see the uses of agentic AI, but the selection of the tool follows the nature of the decision. Generative AI translates natural-language intent into service parameters, predictive machine learning handles pattern recognition and anomaly detection, while agentic AI operates to make and execute decisions within very specific bounds.
“We put the correct AI into the correct components… with very strict guardrails or rules around how you deploy it, how you use it, how you manage it to ensure that it’s trustworthy, to ensure that it’s safe.”
Trust Me, I’m an Agent
How do you make AI “trustworthy” in a network context? Glass’s answer is twofold, and the first lies in the network architecture. Rather than attempting to model the entire network in a single system, TM Forum works with bounded micro-models, each responsible for a defined domain, such as, for example, within the RAN.
“I can be sure that the AI that’s functioning here is safe. It’s not trying to control the world, it’s just controlling this little bit here.” In other words, there are only so many decision options the AI can make.
“That way you’ll be confident, whenever you apply it to your network, will do what you expect it to do.”
Glass didn’t comment on how the different elements are orchestrated in an autonomous network (or, indeed, whether they should be at all – perhaps swarm intelligence works?), so that’s something to check at DTW.
The digital twin is the second safety mechanism. Before any AI-derived reconfiguration reaches a live network, it is tested against a virtual replica.
“I’m not doing something for the first time I’ve ever done it. I’m doing something that actually is a tried and tested pattern.” For critical infrastructure, this is where AI-driven autonomy becomes viable in practice.
