Observed facts: OpenAI’s GPT-Live guide says an app can act while the user is speaking or react to what Live says. It documents both direct native delegation from Live to Decisions and triggering decisions when input or output transcript deltas arrive.
Mechanism: The integration offers two paths: Live calls Decisions directly, or an application observes transcript deltas and uses them as triggers. The two paths place control in different locations: direct delegation lets Live invoke Decisions, while transcript-delta triggers let the application decide how to respond to observed input or output.
Why now: The guide explicitly places actions during speech and at transcript-delta arrival, making the integration relevant to teams building conversational applications that need to respond within an ongoing exchange.
Implications: Builders can assess whether direct delegation or application-managed triggers better fit their control flow. The documentation describes an integration surface, not evidence that either path improves task success or latency.
Evidence boundary: The supplied guide excerpt does not establish general availability, rollout scope, implementation limits, latency, reliability, or production outcomes.
Watch condition: Look for implementation details on trigger semantics, permissions, failure handling, and availability before treating this as a production-ready workflow.