Google prepares Gemini to act inside Android apps to place orders and book rides

Developers who examined Google’s beta app discovered references to a Gemini feature that would perform screen‑automation inside Android apps: simulating taps, filling forms and driving multi‑step workflows to place orders or book rides. The capability appears as an opt‑in Labs‑style option and depends on forthcoming Android platform hooks (noted in the code as tied to a maintenance release) plus app‑level compatibility or certification, suggesting a staged, limited rollout. Implementation notes in the code explicitly warn against relying on the automation for emergencies or entering payment credentials directly in chat sessions; instead payments appear intended to be routed through native app flows or guarded handoffs. Google’s design includes immediate user interrupt controls, telemetry and interaction traces kept for debugging, and a human‑review path for some screenshots or traces to train and moderate the system. Those human review and storage practices echo other Google experiments—Gemini’s web client testing of chat‑history imports and Chrome’s agentic sidebar auto‑browse features—pointing to a cross‑product push to make assistants more agentic. Separately, U.S. Senator Elizabeth Warren has pressed Google for fuller disclosure of how Gemini’s in‑chat commerce and any Universal Commerce Protocol will operate: what signals will be sent to merchants, whether those signals could influence pricing or recommendations, whether advertising or commercial relationships will be flagged, and how user data and intent signals will be routed. Warren set a mid‑February response deadline, compressing the timetable for Google to explain both technical mechanics and privacy controls. The combination of automated in‑app control, telemetry collection and potential backend signal sharing amplifies regulatory and developer concerns: retailers and privacy regulators will demand transparent consent flows, limits on partner access, and technical safeguards (for example, privacy‑preserving APIs or strict operational limits on real‑time intent sharing). Operationally, staged certification and human moderation will slow scale but reduce short‑term risk; product value will hinge on reliable handling of credentials, fraud prevention and consistent behavior across diverse app UIs. Commercially, the Universal Commerce Protocol ambition—to let agents query inventories and complete purchases without redirecting users—promises frictionless buying but risks new forms of opaque personalization or preferential treatment if merchant signals affect ranking or pricing. Whether Google can balance convenience with clear disclosures, robust consent and minimal exposure of sensitive data will determine adoption, regulatory scrutiny and competitive impacts.