ZTE and XLSMART validate a network optimisation AI agent in live wireless operations

ZTE and telecom operator XLSMART have validated a network optimisation AI agent in live wireless operations, in a joint trial aimed at Level 4 (L4) autonomous networks. The results were presented at the ZTE Global Summit & User Congress 2026 in Kuala Lumpur, Malaysia, and reported by Telecoms Tech News on 14 September.

The point of the trial was to run the agent inside production infrastructure rather than a lab or a simulation. The milestone was presented at a joint release ceremony attended by Shurish Subbramaniam, CTO of XLSMART; George Glass, CTO of TM Forum; and James Zhang and Zhang Wanchun, Senior Vice Presidents of ZTE.

Technically, the field trial focused on wireless coverage optimisation, deploying ZTE's Network Optimization Agent to manage operational adjustments across active cell sites. The agent handled the complete operational loop, connecting network perception and analysis directly to automated decision-making and execution, and produced its own optimisation strategies across the trial footprint.

The reported result: across the validated scope, the system recorded an automated closed-loop closure rate of 25 percent. According to project data, the closed-loop operation reduced the manual effort and time that engineering teams normally need to move a network modification from initial analysis to live execution.

For XLSMART, the trial is a step in its migration toward L4 autonomous networks, an architecture defined by automated closed-loop execution and reduced manual intervention. Subbramaniam said the collaboration lets the operator validate high-value agentic use cases directly in the live-network environment and assess their operational effectiveness and measurable value under real network conditions.

ZTE's role is as a technology partner supporting XLSMART's autonomous network roadmap, focusing on implementing autonomous agent software within commercial infrastructure. Following the Kuala Lumpur demonstration, the two companies confirmed they will continue exploring and validating autonomous network scenarios in live-network environments, concentrating future engineering work on enhancing agent capabilities and expanding validated use cases.

The 25 percent closure rate is not full autonomy, but it is a measured result from a commercial network rather than a demo, which moves agentic automation in telecom operations from proof of concept toward partial, quantifiable automation. Telecom networks fit agents unusually well: they are data-rich, rule-bound, and their closed-loop outcomes can be measured in operational metrics.

What to watch next is whether those closure rates scale. The real threshold for L4 is not a single successful scenario but whether automated closure moves from isolated, validated use cases into everyday operation across an operator's commercial footprint, with engineering effort falling as the validated scope widens.