Agentic AI in telecoms networks: operators are moving from detecting faults to fixing them

Telecoms operators have publicly announced more than 1100 AI projects since 2018, and one in four of them now targets the network. The most consequential change in that activity is not its volume but its architecture. AI agents that observe, decide and act on the network, rather than models that produce a recommendation for an engineer, featured in 40% of network AI announcements in the first half of 2026, against 3% in 2024. Operators approaching this as a question of which model to license are solving the wrong problem. The constraint is no longer machine intelligence; it is the orchestration, data and guardrails that allow an agent to act safely on a live network.