Agentic AI is moving into telco operations — but the deployment picture is more uneven than the headlines suggest. STL Partners’ analysis of 64 major announcements and demos from Mobile World Congress shows that live deployments are concentrated in customer care and billing, domains that run on historical data and tolerate batch decision-making. Network operations, where the stakes are higher and the data requirements are more demanding, remains largely in the trial and proof-of-concept stage.
The reason is structural, not technical. Two barriers are slowing real-time agentic AI in networks. The first is inference latency: agent reasoning cycles are too slow for sub-second enforcement at the edge. The second is data currency: even when inference speed is adequate, agents working from near-real-time aggregations cannot safely make decisions because the state they are acting on may already have changed by the time a decision is recorded. Of the 64 announcements tracked, only two live deployments use sub-second real-time data — both in customer-facing fraud detection.
This report from STL Partners, supported by Volt Active Data, examines where agentic AI is actually deployed in telecoms today, what is still in development, and what it will take to extend autonomous decisioning into more critical network operations. Download the full report now.