Deutsche Telekom’s (DT) recent commercial launch of an AI agent platform to help control its radio access network (RAN) was a big step toward an AI future for telecommunication operators, a step Angelo Libertucci, global head of industry for telecom at Google Cloud, said is gaining momentum.

DT’s RAN Guardian Agent uses Google’s Gemini platform to analyze radio access network (RAN) performance, identify anomalies, and autonomously initiate fixes. The initiative was unveiled earlier this year in partnership with Google, and runs on the hyperscaler’s Gemini 2.5 platform, with access to Google’s CloudRun, BigQuery, and Firestore offerings.

DT explained that the analysis involves AI agents continuously scanning publicly available sources, including online directories and social media to identify upcoming public events across Germany. It compiles that list, estimates the scale of the events, determines their location, “classifies them accordingly,” and then the agent vets the data.

Another platform agent takes this insight to determine the network’s ability to handle the expected traffic impact, including nearby antenna capacity and network parameters. It can then recommend optimization options.

The final step uses an additional AI agent to execute approved recommendations, including the ability to reallocate network resources or adjust configurations. It also documents the actions and outcomes as part of ongoing network monitoring and to provide insight into future events.

“Processes that previously took roughly an hour can now be completed within just a few minutes,” DT said of the platform.

Google’s Libertucci noted that, “these are agents talking to agents providing context, reasoning, recommending configuration changes, autonomously, making those changes without humans.”

DT’s steps are being echoed across the telecom space, with vendors and operators increasingly tapping AI platforms to help monitor network performance.

A recent Google Cloud report found that more than half of telecom executives surveyed said their organizations were actively using AI agents in production, with 43% noting they were using 10 or more in those environments. More importantly, two-thirds stated they were already seeing a return on that deployment investment.

AI flexibility in the network

The depth of AI-infused telecom plans highlight the need for AI platform providers to remain flexible in terms of their offerings. Libertucci explained that for Google Cloud, the key is to support operators at various points in their AI journey.

“They might not have fully autonomous, no human-in-the-loop yet, like they're all on different journeys and trust levels,” Libertucci said of the diverse carrier landscape. “I think human-in-the-loop kind of gives them still that feeling, that they don't want to kind of totally give them to an agent yet, but I think over time, once the agent makes a recommendation and you push the change, and the human certifies the change, and it's correct, and you're happy with the result, you do that over and over and over again, you kind of build this trust level.”

Libertucci added that the cloud giant has, “earned, maybe, the trust of carriers because they know that we run and operate the world's largest private network ourselves.” This includes Google Cloud’s more than two million miles of managed fiber, 200 points of presence (PoPs), and 40 regions to support billions of users.

“We do this with an incredible amount of automation and very little operational kind-of-overhead,” Libertucci said, adding that, “the number of engineers that we do this with are a fraction of the people that a regional telco would have in the United States, for example. So we have this very prescriptive point of view about leveraging the products and services that we use ourselves.”

Need to avoid fragmentation, gain trust

These AI-infused efforts have also bolstered a focus around driving AI standards. Some of this initial work has solidified at trade groups like the AI-RAN Alliance, which formed early last year in an attempt to steer the use of AI into RANs for better performance, lower operating costs, greater efficiencies, and to support new business models.

Google Cloud is not a current member of that specific group, but Libertucci did note that the cloud giant works with others like TM Forum and directly with telecom vendors such as Ericsson, Nokia, and Amdocs.

These trade groups or organizations can bring what Libertucci said was awareness to challenges that are especially important to the AI space “because everyone is moving at such speed, and the pace of innovation is moving so quickly, and everyone has a project that they think can provide immediate value for the organization. But if everyone kind of goes off and does their own thing without any structure or enterprise-wide framework, then I think we end up with the same problem.”

“What we learned from history is that we don't learn from history,” Libertucci said, adding, “what we don't want to do is go from fragmented technology stacks to fragmented AI stacks. We don't want to make that same mistake.”

While there is progress toward avoiding fragmentation, a bigger hurdle could be in getting more operators to entrust their network data to AI models.

AI platforms generally perform better when they have as large of a data set to train from. However, some operators have openly touted the “private” nature of their cloud operations, as opposed to others that have been more open to embrace the public cloud.

Will 6G be the answer?

Libertucci’s view is that this hurdle will be worked through as the industry moves toward a 6G standard that will delineate the need for where AI should be used in a network deployment.

“If you look at 3GPP, there's clearly a requirement for inferencing at some parts of the network,” Libertucci said. “Over time, we'll debate on where the edge is and where that needs to be done. Obviously, Nvidia, Nokia feel that needs to be right out at the RAN site.

“I'm not sure the world's going to settle on that, but at some point there's going to be some non-cloud region inferencing, and it'll be closer to the edge of the network, and that'll be debatable. But certainly the leveraging of AI to build an autonomous networking framework is an initiative that we're working on today in an agentic fashion – as we talked about with Deutsche Telekom, with humans out of the loop in production – I think this is a huge testament to what's capable.”