Three months into Claudee for Desktop](/tools/claude-for-desktop/)'s voice features being available, someone ran Claude as an active meeting participant on Zoom - not lurking, not transcribing, but answering direct questions from four different attendees in real time. The whole thing held up without glitches.
The setup didn't use a phone or a third-party Zoom integration. The person ran Claude for Desktop on a computer and routed its audio directly into the call. Claude kept pace with questions from different speakers and maintained context throughout the session - the harder part when multiple people speak in sequence.
What stands out isn't the technical side - it's that no one has productized this use case. The closest existing tools are AI note-takers that listen and summarize after the fact. Running a model as an active voice respondent is different: it's participating, not observing. Obvious first applications include a standing research resource during client discovery calls, an AI that can field product questions during a sales demo, or real-time fact-checking during presentations.
One consideration for any professional deployment: several U.S. states require all-party consent before recording a conversation, and an AI speaking in a meeting raises the same legal questions as a recording device. Explicit disclosure to all participants is the safe move before taking this anywhere near a client call.