AI Agent Use Cases

Added today

10All 231

1–20 of 231

Dev

An agent that builds a live voice conversation interface

A case of reading official documentation to implement a web interface for voice conversations with Gemini.

How it works

  1. Simon Willison gave documentation to GPT-6 Astra Extra High and had it build a web interface for testing Gemini Live.
  2. The published interface lets users choose a model and voice, optionally enter a system prompt, and start a voice conversation.
  3. Users can interrupt the model while it is speaking; the implementation uses WebSocket and the Web Audio API.

Results

The author published a browser tool for voice conversations and viewing conversation transcripts.

Notes

The author says it was implemented without libraries. The interface warns that transcripts may include speech interrupted before it was played.

Source

Simon Willison Original post ↗

Example prompt · Codex

Please implement a web interface in [development environment] for testing Gemini Live voice conversations. First check the current official documentation for [model to use], access permissions, authentication, and billing conditions; do not create a new paid connection without approval. Include model and voice selection, optional system prompt input, microphone start and stop, and conversation transcripts. Implement a WebSocket connection and audio input/output using the Web Audio API, stopping the playing response when the user starts speaking. Provide runnable code and setup instructions. Do not leave credentials in code or logs. Use [test sentences] to check connection failures, denied microphone permission, interruptions during responses, and session termination. Clearly distinguish interrupted speech that remains in the transcript.