Omdia spoke with Deutsche Telekom about its recently launched RAN Guardian Agent solution.
Omdia view
Summary
No doubt agentic AI is the hottest new technology for telecoms and beyond, but so far, it has primarily resided in PoC (proof-of-concept) city. Omdia was encouraged, then, to see that Deutsche Telekom (DT) has launched its RAN Guardian Agent solution into production. Omdia recently spoke with Ahmed Hafez, Senior Vice-President for Tech Strategy and Data and AI, to learn more from an early adopter pushing forward with a nascent but potentially transformational technology.
DT went big: multi-agents for RAN Optimization
DT has been working on AI for some time, exploring developments across the ecosystem. The development team decided to focus on concrete actions and implementations, starting with exploration use cases a couple of years ago. They realized there was a significant gap between proof of concept and implementation, with implementation requiring four to five times more effort.
As it was early days, the team faced skepticism about agentic AI. However, DT prides itself on a “learning by doing” approach, believing that working with live systems is the only way to truly understand new technology, and that PowerPoint presentations are insufficient. They decided to implement not just agents, but multi-agentic systems right away, a potentially risky approach, but one they felt needed to happen to have the most impact.
The team decided to focus on Radio Access Network (RAN) optimization due to its high impact on customer experience. They aimed to address the daily problem of network tuning for the best customer experience, especially during congestion. Previously, a team tried to manually manage about 1,000 events (concerts, sports matches, etc.) across Germany annually. Clearly, their ability to support them was limited by human capacity, as tuning can take an hour per site.
They split tasks across multiple agents, allowing them to manage their actions and scope of influence more closely. DT estimates that 75% of actions are fully autonomous, meaning they can be carried out without human approval, while the other 25% need a human to put the agent’s recommendation into action. Humans have visibility into everything the agent is doing, and those agents act based on declarative instructions. Intent-based operations, which could cause conflicts between agents, aren’t quite ready for production yet, in DT’s opinion. The three agents involved are:
- Event agent: Gathers information about events from various sources (news, social media, etc.), verifies the information, and assigns confidence scores.
- Monitoring agent: Recontextualizes alarms, evaluates capacity needs, and prepares for potential issues.
- Remediation agent: Changes network configurations in real-time as events build up, taking about a minute to implement changes compared to an hour previously.
After running the event agent, DT identified 40,000 relevant events that occur annually in Germany, providing further impetus for automating network tuning.
It’s hard work, but worth it in the end
Ahmed Hafez conveyed the significant effort and brainpower needed to get RAN Guardian Agent into operations. They had to create guidelines on managing AI and agentic AI, since it’s a new system with its own challenges. As it turns out, DT needed to track two sets of KPIs:
- One was the existing network performance KPIs (throughput, latency, etc.)
- The other was a new set of KPIs specific to agent performance, such as response time and accuracy.
DT continuously monitors the agents to prevent drift and ensure performance.
Hafez is confident that the agentic AI work done for RAN optimization will transfer to other network domains. In reflecting on his discussions with other CSPs considering agentic AI, he found a significant gap in understanding. He reports that some are overly skeptical, while others are excessively optimistic, because no one fully understands what it takes to make agentic AI work until they actually start working with the systems.
He agrees with Omdia that when used as part of a re-thinking of network operations, agentic AI is truly a transformational technology. It remains early days, but he is confident RAN Guardian Agent will generate a meaningful ROI in due course.
Appendix
Further reading
2026 Trends to Watch: Telco AI Software (November 2025)
Agentic AI: An Evolution with Transformative Potential for Telecom Operations (October 2025)
Separating the Hype from the Reality of Agentic AI in Telecom Network Automation (October 2025)
Author
Roz Roseboro, Senior Principal Analyst, Telco Software and AI
