An AI system now claims it can replace the person standing at the front of the room clicking through a software demo.
Researchers introduced Rhetor, a multi-agent system that takes a running web application and its source-code repository and produces a scripted live demonstration — complete with synchronized narration and real-time voice question answering. The system explores the UI, maps it against source code, writes a script constrained to elements it has actually seen, rehearses that script in a loop until it converges, and then runs the whole thing live. Across six test sessions on four applications, including the open-source whiteboard tool Excalidraw, Rhetor's internal locator-firing rate hit approximately 0.92 on a 53-action workload and reached a perfect 1.00 on the public reference app after two rehearsal iterations.
The gap Rhetor targets is real: pre-recorded demo videos go stale the moment a UI changes, and general-purpose browser agents are not built to narrate coherently or handle audience questions. A system that can do all three simultaneously would cut a genuine recurring cost for software sales and developer-relations teams. The voice question-answering piece is the hardest claim to verify from a paper alone.
The researchers also propose a ten-metric benchmark across six application categories — a sign they know a case study on four apps is not a proof of generalizability. Whether Rhetor holds up outside a controlled lab setting, against real enterprise UIs and unpredictable audience questions, is the part the paper cannot yet answer.