Multi-agent AI environments and the importance of trust

Interview

Netcracker’s Head of Strategy and Marketing, Sue White, discusses why telcos’ rapid adoption of Agentic AI must not risk compromising their operational integrity

As expected, the evolution of AI was one of the major themes at this year’s DTW conference in Copenhagen, with discussions moving swiftly from Generative AI to Agentic AI. This change – allowing AI agents to go beyond recommending solutions to taking action autonomously – potentially represents a major paradigm shift for how telcos operate, but only if they can embrace the technology at scale.

But while echoes of the ‘move fast and break things’ philosophy of Big Tech evangelists were certainly present in these AI discussions, so too were voices promoting a more measured approach – after all, an AI agent’s mistake within a telco network could potentially impact millions of customers.

Sue White, Netcracker’s Head of Strategy and Marketing, was one such voice.

“Ultimately, you have to be disciplined and balance autonomy with oversight,” said White, speaking at the event. “That doesn’t mean you can’t move with speed, but it means you don’t let AI Agents loose on areas of your business that are mission critical until you’re certain they will deliver the correct responses. And you have to ensure you’re protecting the rights and identity of your customers throughout the process.”

Step by step: Building a multi-agent framework

For White, the implementation of Agentic AI naturally begins with simpler tasks within the business, with the AI agent given access only to specific tools, data pools and with limited ability to action requests.

“What we’re deploying first are agents that have a fairly limited scope to ensure the tasks they do are correct and the agents are stable,” said White. “You can begin with a high level of human oversight and then reduce this oversight as you build up trust. If you make the use case too broad initially, there is more chance for the LLM to hallucinate.”

From here, the number of AI agents can be steadily increased across the business, each specialised around a specific part the business, from agents assisting call centre operators to those supporting engineers during network outages. The result is a broad tapestry of AI agents, each trusted to act within a limited remit.

“You might have a billing agent that is able to understand everything about a customer’s bill, another agent that can look into users’ tariff plans, another for troubleshooting, and then a master agent that coordinates between them,” explained White. “Initially it’s about having agents across the business – whether it’s customer-facing or internal – that solve a limited number of tasks and then allowing these agents to talk to each other for more complex use cases.”

“Eventually, these agents will start peering with each other, but we need good governance in place for that,” she added.

Checks and balances

Of course, orchestrating a myriad of bespoke agents comes with significant complexity. Each agent is potentially drawing data from different parts of the telco business and leveraging different AI models, potentially creating challenges around data privacy and security.

“It is absolutely crucial that we do not to give public models access to proprietary telco systems,” said White. “This is where new frameworks and technologies are coming in place, like the MCP (model context protocol).”

The MCP is an open standard, open-source framework introduced by Anthropic in November 2024 to standardize the way AI integrate and share data with external tools, systems, and data sources. It includes various specifications for data sharing across different platforms, allowing enterprises to deploy local MCP servers to act as a secure intermediary between AI models and secure data.

“This becomes incredibly important in the business because you’re basically grounding the model in the information it can access,” explained White. “You’re saying for this particular request, here’s the instructions and here’s a way get this data that will allow you to solve the problem, but you’re only allowed to use these selected systems.”

In addition to building frameworks for how these agents interact with data and with each other, White also stressed that telcos need to be careful with exactly how their customers’ data is exposed.

“You have to build in access controls. You have to obfuscate any information that’s sensitive to the customer, so if an agent gets billing data, for example, it can’t be tied to the customer,” she said.

Part of the challenge here is the sheer breadth of data available to telcos, with many struggling to successfully pull together data from many disparate sources. Perhaps ironically, AI itself could be the solution here too, cleaning up messy data, formatting it for easier access by agents, and anonymising sensitive customer information.

Enormous potential

While it is undeniably still early in the telecoms industry’s journey of Agentic AI adoption, real-world examples of its use are already emerging, generating major efficiencies as a result.

“We’re already seeing agents being used on both the front end and the back end of telco businesses. In the case one of our US customers, they have a digital assistant that can query really complex invoices with thousands and thousands of lines. These documents would have would have taken people days to sift through and compare, but with our AI agents it can be done very quickly,” White explained. “Equally, we have agents that can take a complex business requirement document and build a new offer for a B2B customer. It will ask you for any information its missing and then go and configure this new offer in the catalogue. You can do in an instant something that would have taken weeks to gain approval.”

As telcos accelerate their adoption of Agentic AI, the path forward demands a delicate balance between innovation and responsibility. By building trust through careful, stepwise implementation—starting with limited scopes and strong human oversight—telcos can harness AI’s transformative potential without compromising customer privacy or network integrity. Ultimately, the future belongs to those operators who embrace AI not just as a tool for efficiency, but as a strategic partner built on transparency and trust.

For more information about Netcracker’s Agentic AI capabilities, visit the website

Also in the news:
US judge rules Huawei must face charges of fraud and racketeering
Optus ditches football rights to focus on telecoms
Nokia launches digital twin platform Enscryb to digitalise energy sector

Author: Ernestro Casas -

This post was originally published on this site