In 2025, China Mobile tested a fully level 4 autonomous network in Guangdong, and the results were certainly striking. Fault diagnosis time cut from 44 minutes to roughly four; mean time to repair reduced by 50%; an 80% reduction in back-end troubleshooting hours; and a “dark NOC” without anybody sitting in it.
“That’s 10 times the size of BT. They’ve rolled it out to 300 million customers,” explained George Glass, CTO of TM Forum, in an interview with TelcoForge.
Yet, only around 6% of operators globally have reached this level of deployment.
The gap between headline results and industry-wide adoption is partly a complexity problem – and we’ll come back to that, absolutely.
But Glass’s account of organisational resistance is more revealing. He expected friction at the traditional fault line between IT and network operations. “I spent 30 years in BT, and IT and network were separate organisations within the organisation. And that I see that mirrored across the globe.”
What he did not expect was how fragmented the network function itself would prove to be.
“The IP team don’t talk to the core team, don’t talk to the access team because they’re different technologies, they’re different vendors.”
When TM Forum began building cross-domain fault management spanning IP, transport, and access, it found that roughly 80% of the underlying process was common across domains, with 10-20% genuinely domain-specific. However, standardising the common elements, which sounds common-sense, turned out to be disruptive.
“We’re starting to uncover, even within the network side of a CSP, silos of individuals who for the last 20 years, they were the guru, the expert in a particular area. And now we’re abstracting that.”
Should we think of this as a Luddites’ last stand? Probably not. In fairness, even where the technology is similar, the business processes, supplier relationships and the kinds of detailed knowledge are quite different. People have become specialists, but largely from necessity.
“If you’ve got Ericsson providing one area and Nokia providing the other, the likelihood of somebody being an expert across both? Very slim. You’re either an Ericsson or a Nokia person,” said Glass.
Bro Down
This adds a layer of complexity to another conversation, which the TM Forum executive tackles head-on.
“Nobody’s going to go on stage at DTW and say, ‘Brilliant, we can get rid of 20% of our workforce.’ But ultimately, that’s what the shareholders are looking for because they want the cost driven down.”
China Mobile’s deployment has, by publicly available figures, displaced the equivalent of 5,500 people’s work in fault management.
“It’s happening,” Glass said. “Nobody’s going to go on stage and say it. But it’s happening.”
There are a few ways to parse this information. It will please shareholders in the short term because “number go up”. Meanwhile, the average age of telecom engineers is rising globally, so before long, there will be fewer and fewer skilled engineers as retirement hits.
Do the job losses caused by automation and autonomy outpace that organic decline? TelcoForge hasn’t seen conclusive information either way.
In the longer term, though, there’s a more insidious problem emerging. Running an autonomous network still requires people, of course, as Glass points out keenly.
“Somebody still has to train the model, somebody still has to test the models, people have to become familiar with the AI, need to be confident that the AI is in fact trustworthy. So there are different skills required.”
Guangdong’s dark NOC does not run itself entirely; it runs itself within parameters set and monitored by people who understand both what the technology does and what it is supposed to achieve.
It requires people who understand both the networks and the AI to be sure they’re working as required, and also the business processes and supplier relationships involved in dealing with any problems. Autonomous operation can mask a huge amount of network complexity, but somebody still needs to understand it.
Back to the Basics
The question is – who will that be in future and how will they learn it? The big vendors would, undoubtedly, be very happy to take on the challenge for telcos, but that leaves them locked in by the difficulty of knowledge transfer even if open networking takes off.
Building institutional knowledge sounds like a good idea, in principle, but that requires a few elements to be in place. The ability to codify knowledge within the organisation is under pressure if layoffs are on the cards.
What’s more, experienced engineers carry judgment in context as well as technical knowledge: which vendor’s documentation can be trusted, what internal processes can be streamlined under what circumstances, and so on. That knowledge is not easily codified, but is hugely important for dealing with both technology and the usually messier bit to do with people and business processes.
In this situation, is there a case for some different approaches to “human resources”? Should we perhaps be bringing on promising young interns with the promise of a job for life, with loyalty from the organisation to reward loyalty from the staff? It’s quite a 1950s – or maybe quite Japanese – concept, but it may be the best way to bring in and hold on to the relatively few, increasingly crucial people who can keep the lights on in any given network.
The promise of autonomous networks is undeniable, and, at DTW, we’ll see more demonstrations of how far the technology has come. But the path from 6% adoption to industry-wide transformation runs through problems that are organisational and human as much as technical.
Glass himself acknowledges it ruefully. “If it was easy, somebody would have done it before. It’s a journey that we’ve been on for four or five years. We’re starting to get momentum around it. We’re starting to get traction in the marketplace.”
