AI Agent Bobbi Resolves UK Police Calls

PYMNTS reports that an artificial intelligence agent called Bobbi resolved 82% of citizen queries in its first week live across three police forces in the United Kingdom, and that none of those interactions were passed to a human agent. PYMNTS also reports that UiPath, in a May 5 press release, announced agentic capabilities inside its platform tailored for government and regulated industries and said the update lets agencies run AI models inside their own infrastructure, including cloud-hosted or fully self-hosted options. The release included a quote from UiPath's Public Sector Chief Technology Officer: "As government agencies adopt agentic automation, our focus is helping them stay in control of their data, models, and how AI is used," PYMNTS reports. PYMNTS further reports Google announced new tools for institutions, an open model option, and that its models now process 16 billion tokens per minute via its direct API, supported by new eighth-generation TPUs.

Editorial analysis - technical context: The reported combination of on-premise/self-hosted model support and data-sovereignty controls addresses a common procurement barrier for regulated agencies. For practitioners, deploying agents without cloud egress typically requires containerized model serving, orchestration across existing infrastructure such as AWS and OpenShift (PYMNTS reports UiPath's platform runs across those environments), and integration hooks to legacy telephony and case-management systems.

Public reporting frames these items as part of a broader vendor push to make agentic automation acceptable for regulated institutions. High-resolution early deployments like the reported Bobbi rollout provide a real-world confidence signal for contact-center use cases where automation must meet privacy and compliance constraints. Vendor claims of high token throughput and new TPU generations signal ongoing investment in both model capacity and deployment tooling at scale, which matters for latency-sensitive voice interactions.

For practitioners: monitor independent evaluations of resolution quality and escalation safety from pilot deployments, telemetry on failure modes in voice-to-intent pipelines, and procurement language that codifies acceptable hosting models and data controls. Observers should also watch whether independent audits or publishable metrics accompany vendor performance claims reported by PYMNTS.