Google announces models able to continue a conversation while carrying out tasks in the background. This fluidity makes it necessary to make the shift from speech to action more legible.
A conversation can carry on while a search or an operation runs elsewhere. This is one of the capabilities highlighted by Google for Gemini 3.8 Live and its Extended Thinking version, presented on 15 September 2026.
Speaking and executing in parallel
According to Google, the models can call on tools in the background while keeping the exchange going. The Live version also processes visual context in near real time and switches between ninety-seven supported languages. Extended Thinking is aimed at tasks requiring more complex reasoning.
These capabilities are announced by the provider. Their effectiveness in a particular situation must be assessed on that situation, rather than inferred from a demonstration or a ranking.
The moment speech commits
For reception, training or accompanying a visitor, a voice that does not pause every sentence to consult a tool may feel more natural. This continuity nonetheless presents a design difficulty: the user must know whether the agent is explaining, searching, preparing or carrying out an operation.
In a house, confirming an appointment, changing a piece of information and proposing an option are different acts. A voice interface should name them clearly and provide suitable validation when the decision commits the customer or the company.
The useful test is therefore not limited to the quality of the voice. It also concerns interruptions, ambiguities and human takeover: what happens if the person changes their mind when a task has already begun?
Fluidity can improve the experience if it keeps responsibilities visible. The service must keep an understandable record of what was requested, authorised and actually carried out. A successful conversation does not, on its own, prove that an action was correct.
Cette publication est également disponible en :
