AI agents on the DaasLabs Data Fabric Framework do the routine work end to end across network operations, fraud and revenue assurance, customer care, churn, BSS/OSS and billing, enterprise, regulatory, wholesale and the data platform. Policy decides what goes straight through, and people approve the exceptions.
Roster figures count the agents on this page. Live figures come from the DaasLabs Telecom demo platform, not from client operations.
| Case | Fraud type | Risk | Decision | Decided by | First evidence |
|---|---|---|---|---|---|
| Loading… | |||||
Each agent works on governed BSS, OSS and CDR data, calls only the tools it is allowed and drafts an outcome with its evidence.
Risk tier, exposure limits and the agent's autonomy level decide: straight through, or to a person.
A named owner reviews, edits or rejects only the cases that need judgement: blocks, credits, claims and anything a regulator sees.
Every signal, tool call, decision and action is kept as an audit trail; outcomes feed the models. See governance →
Every agent has a named human owner, an autonomy level set by governance, a list of tools it may use and a stated line it never crosses without a person. Agents marked in demo run inside a DaasLabs Telecom accelerator; blueprint agents are reference designs we configure to each operator's systems and controls.
Every agent starts at the level governance agrees with risk, privacy and the business owner, and moves up only when its override rate and evaluations support it. The bars show how many agents on this page run at each level.
Every agent has an accountable owner who approves, edits or rejects its exceptions. Blocks on enterprise lines, credits and partner claims always need a person.
Risk tier, exposure and adjustment limits decide what goes straight through; limits are configuration agreed with each operator's risk team, not code.
Agents call typed, limited, reversible actions through the Action Fabric — never free-form changes to billing, network or subscriber records.
Subscriber identifiers are masked in prompts and screens; agents see only the data the case owner is entitled to, inside the operator's own tenancy.
Every signal, plan, tool call, decision and action is logged with lineage back to the source record, ready for auditors and regulators.
One control pauses an agent, a squad or the whole workforce; work in flight returns to the human queue.