Login
Register
Dawn Bushaus
Telcos need to consider the impact of agentic AI on humans
Nathan Bell, Digital Partner at Kearney and the former Chief Digital Officer of Singapore’s M1, sees some uncertainty and disillusionment in the telecoms industry. And while he thinks AI has transformative potential, particularly in telco networks, he says the experience of fintech startup Klarna should serve as a cautionary tale when it comes to agentic AI. This article is the first of two based on an interview with Bell in the lead up to DTW Ignite 2025. He will participate in two AI panel discussions at the event.
In February 2024, Swedish payment startup Klarna launched an AI assistant to replace customer service reps, handling 2.3 million chats in its first month – two-thirds of all customer interactions. By December, CEO Sebastian Siemiatkowski declared AI could do all human jobs and confirmed that Klarna had stopped hiring.
Fast forward five short months and Siemiatkowski has done a complete about-face. After customers complained about Klarna’s lack of empathy and inaccurate responses, Siemiatkowski admitted to Bloomberg in May that the company overused its AI agent, and he acknowledged that prioritizing cost led to lower quality.
CSPs should heed Klarna’s learnings as an early adopter, says Bell, adding that in his consulting work he is starting to hear about examples of AI not measuring up to communications service providers’ (CSPs’) expectations.
“If you combine the current geopolitical uncertainty [around tariffs] and the white-horse hope that we had for AI, I think people are getting a bit nervous,” he says. “While it’s important to be bold with your ambition, defining metrics that help validate how a use case is progressing is important to ensure there is an ability to pivot or pause as required to maximize business value.”
Bell believes the use of AI – particularly generative AI (GenAI) and agentic AI – in networks could fundamentally transform network engineering. Indeed, the goal of agentic AI in the network is to enable autonomous, intelligent decision-making across operations, reducing (or eliminating) manual intervention.
Unlike traditional AI models that rely on pre-defined rules, agentic AI systems learn, reason and act independently. The potential is both exciting and frightening as operators begin to experiment with individual use cases focusing on network planning, monitoring, restoration and assurance.
“If you were to stitch all of these great use cases together, you can start to reimagine what network engineering is going to look like over the next five years,” says Bell. “And I think that’s quite scary for a lot of people, especially in a function that has seen evolution in new tools and capability. But this represents a true revolution of their function.”
Bell points to a TM Forum Moonshot Catalyst, to be demonstrated at DTW, showing a “Concierge AI Agent”, which is a proactive, intelligent assistant that continuously monitors the customer journey and engages before problems arise. The agent is supported by an AI ecosystem where Open Digital Architecture (ODA) components communicate via Open APIs and natural language protocols.
When a service issue or anomaly occurs, the Concierge AI is alerted. It gathers context from billing, network and service systems, consulting with other AI agents in real time. Then, it reaches out to the customer proactively with solutions to issues and personalized insights. The overall goal of the project is to enhance ODA standards so that they can support AI-to-AI interoperability.
“An engineer would normally take two or three hours to go and check and validate all of these aspects as they try to distinguish between symptom and root cause that, while this is doing it in seconds,” says Bell. “Network engineers have never been as disrupted as this. I think that’s why it’s the human factor with human-driven processes that’s going to potentially slow down the broader adoption of AI for networks more than anything else.”
Some telco C-level executives are overlooking the impact of the human factor when it comes to AI adoption, according to Bell. He offers an anecdote of one CEO’s AI showcase for employees that backfired when viewed through this human lens. The idea was to excite and energize employees with a presentation about all the different AI use cases the company was deploying and how it would impact different functions.
Bell sat quietly at the back of the room observing employees’ reactions. When the presentation was over, he told the CEO he didn’t think it landed well. But the CEO insisted it would have the desired effect.
A week later, Bell spoke with the CEO again and asked about the showcase: “He said: ‘I don't really want to talk about it’,” Bell recounts. As it turns out, employees were alarmed by the demos.
“Sure enough, HR got hit with a whole bunch of emails, calls and people dropping by to ask, ‘So, when do I get my [severance] package? ...I’m clearly not going to have a job in the next 12 to 18 months’,” says Bell. “So, I think there’s a massive disconnect in that regard.”
AI companies ‘want their money back’
Bell is also seeing some disillusionment with AI because of soaring costs. He points to one large operator he’s advising that has many AI use cases running concurrently.
“None of them are paying back,” says Bell. “And they’re really worried because they’ve committed to a three-year contract to go and build and run this stuff. So, now they’re asking themselves: ‘Were we too eager? Did we sign these contracts without understanding what the implications of costs were going to be?’”
The issue is one of licensing. AI software companies “are getting more sophisticated”, according to Bell, and they are starting to price AI agents as if they are people, each requiring its own license. “So, all these changing commercial models are starting to be played on the commercials,” he explains. “And then people will say to you that surely with the cost of compute getting cheaper, AI will get cheaper.”
While that might be true in principle, “people have invested hundreds of billions of dollars in these systems,” Bell says. “They want their money back, and software providers need to show a clear ROI for their investments.”
By some estimates, there are as many as 70,000 active AI companies. “That’s a lot of investors who are saying, ‘Okay, how do I make sure I get my money back versus someone else?’”
In the next article, Bell discusses why CSPs must learn from the past to succeed with AI transformation. He also highlights the AI use cases with the best potential to deliver return on investment.
