Simon Willison had Claude write code to find a mystery beep

Simon Willison heard an intermittent beep in his home and could not identify the source. He asked Claude, Anthropic's AI assistant, to help. Claude wrote a Python script that used the computer's microphone to record audio, detect beep-like sounds, and log the time and volume of each one. Willison ran the script and left it recording in the background.

The script caught the beep. The log pointed to a specific time, which let Willison narrow down which device was responsible. The agent did not just answer a question. It designed and built a sensing tool fitted to the specific problem, ran that tool, and returned evidence a human could act on.

This is a pattern worth naming for product teams. The agent extended its own reach into the physical world by writing instrumentation code on demand. Willison did not specify how to detect the beep. He described the problem, and the agent chose the method. The human contribution was framing the problem and deciding to trust the output. The agent's contribution was the design and execution of a novel data-collection approach.