Voice-to-Action Interfaces
Voice-to-action interfaces are product experiences where spoken intent directly drives software actions, tool calls, UI changes, and external-system updates rather than only producing a conversational answer.
Key points
- OpenAI's GPT Realtime 2 e-commerce demo shows the pattern clearly: the user asks verbally, the agent searches products, reads reviews, checks weather, compares constraints, opens pages, adds items to the cart, and updates the visible UI [src-083].
- The analytics demo applies the same pattern to business intelligence: the user asks for filters and root-cause analysis, the voice agent manipulates the dashboard, launches an investigation, and returns a concise explanation when asked [src-083].
- The interface contract is different from chat. The user is not asking for a paragraph; they are delegating an operation across application state, tools, and external context [src-083].
- Good voice-to-action agents must decide when to speak, when to stay silent, when to use a preamble, and when to update the screen instead of narrating every step [src-083].
- The pattern depends on structured tool access and state awareness: the model needs to inspect page state, choose among tools, remember the user's plan, and avoid losing context across multiple turns [src-083].
Related entities
Related concepts
Source references
- [src-083] OpenAI – "Build Hour: GPT-Realtime-2" (2026-05-13)