Deutsche Telekom and Google Cloud Deploy AI Agent to Modernize 5G Operations During Calls

The AI agent, built on Google Cloud’s Vertex AI platform, sits between Deutsche Telekom’s network‑management console and its 5G radio units, intercepting operator commands in real time and translating them into the proprietary configuration language used by the base‑band software. According to CDO Magazine, the system can ingest a technician’s spoken request—such as “increase uplink bandwidth on sector 12‑A”—and automatically generate the corresponding Netconf or REST‑CONF payload, push it to the relevant gNodeB, and verify the change without human intervention. The automation loop runs in under two seconds, a dramatic speedup over the manual “call‑center‑style” troubleshooting that previously required operators to log into a separate GUI, copy‑paste commands, and wait for confirmation.

Google’s involvement goes beyond the AI model itself. VentureBeat notes that the partnership leverages T‑Systems, Google’s German subsidiary, to host the agent in a “sovereign cloud” that complies with GDPR and German data‑sovereignty requirements. By keeping all telemetry and voice data within the EU‑based infrastructure, Deutsche Telekom avoids the cross‑border data flows that have hampered other AI deployments in the region. The sovereign cloud also provides dedicated networking paths to the carrier’s edge data centers, reducing latency for the bidirectional streaming of audio and command responses—a critical factor for maintaining the sub‑second reaction times promised by the system.

The agent’s speech‑to‑text front end is powered by Google’s Speech‑to‑Text API, while the natural‑language understanding component draws on a fine‑tuned PaLM model that has been trained on Telekom’s internal operation manuals and 5G performance logs. CDO Magazine reports that the model was exposed to more than 10 petabytes of historical network data to learn the nuanced phrasing operators use when diagnosing issues. This extensive pre‑training enables the agent to disambiguate homonyms (“cell” vs. “sell”) and to handle multilingual commands, reflecting the multilingual environment of German telecom operations.

Deutsche Telekom plans to roll the AI assistant out across its European 5G footprint over the next twelve months, starting with the high‑traffic urban cores where manual adjustments are most frequent. The company expects the automation to cut average resolution time from the current 15‑minute window to under 30 seconds, according to internal projections shared with CDO Magazine. In addition to speed, the system is designed to capture a structured log of each interaction, feeding back into the training pipeline to improve future accuracy—a continuous‑learning loop that mirrors the approach Google has taken with its own cloud AI services.

While the deployment marks a significant step toward fully autonomous network management, analysts caution that the technology still relies on human oversight for safety‑critical changes. As noted by VentureBeat, the sovereign cloud architecture includes a “human‑in‑the‑loop” checkpoint that requires an operator to approve any configuration that could affect service‑level agreements or regulatory compliance. This safeguard reflects industry best practices for AI‑assisted control planes, ensuring that the AI agent augments rather than replaces skilled technicians.

The initiative also dovetails with Deutsche Telekom’s broader AI strategy, which includes the Magenta AI Call Assistant unveiled at Mobile World Congress 2026. That service, built with ElevenLabs’ voice‑cloning technology, demonstrates the carrier’s commitment to embedding conversational AI across both customer‑facing and back‑office functions. By unifying the same Google Cloud infrastructure for both the call‑assistant and the 5G operations agent, Telekom can leverage shared data pipelines and security controls, creating a cohesive AI ecosystem that spans the entire value chain.