Start with work that can be checked
I spent years at ETI Software Solutions working with the systems ISP operators use to provision service, investigate faults and prepare compliance data. Those workflows give XSI a specific place to apply agentic AI.
An OLT investigation can require current alarms, optical levels, affected subscriber equipment and a documented diagnostic sequence. The sequence varies by device and fault, but its inputs and permitted actions can be defined.
BDC preparation offers another structured workflow: reconcile locations, verify availability records, apply a submission schema and review exceptions. A file passing validation still needs accurate underlying data and the required certification.
Keep the necessary capability close to the network
Fault investigation can become harder when WAN connectivity is degraded. A local model can continue the parts of the job whose data and tools remain available on the operator’s network.
Subscriber and network information also needs explicit access and transfer controls. Local inference is one way to reduce external processing. It does not remove privacy obligations, and cloud processing is not categorically prohibited for every ISP workflow.
XSI LodeStone’s approach combines local inference with governed integrations. The operator determines which connections are available and what information an agent may send through them.
Integrate with the operator’s actual systems
An ISP may run several generations of access equipment alongside its OSS/BSS and GIS platforms. An agent needs the relevant identifiers, current device state and supported operations from those systems.
TMF interfaces and vendor adapters can reduce translation work, but implementations still differ. Authentication, API versions, data quality and deployment-specific behavior have to be checked.
The value of a curated skill lies in this operational detail: the source it uses, the action it permits, the evidence it produces and the conditions that send work to a person.
Measure the first deployment
A useful initial scope is a defined job such as preparing a diagnostic recommendation or reconciling one class of reporting exceptions. Shadow operation can collect proposals and outcomes before an operator grants write authority.
Measures such as operator review time, incorrect recommendations, escalation quality and total cost per completed job give the deployment a concrete record. They also show which tasks should remain under close human supervision.
ISPs have industry associations, common vendors and established operational communities through which deployment experience can travel. XSI’s focus is building useful work into that environment, with the controls and integrations the operator can inspect.
