Claude Code-built Beep Locator finds beeping device within 40cm

Willison's project began with a frustrating problem: an intermittent beep in a cluttered box room. Rather than unpack every box, he turned to Claude Code, an AI coding agent, to build a solution. The agent researched the problem and surfaced a 2007 academic paper, BeepBeep, which described a way to use commodity devices—speakers and microphones—to range and localize sounds. That research directly informed the design of Beep Locator, which coordinates multiple listening devices via a Python web server and WebSockets.

The system's development also showcased a practical technique: giving the AI agent the ability to perceive audio. Claude lacks native audio input, but by running JavaScript and Python code that captured and analyzed microphone data, Willison effectively gave it 'ears.' This let the agent iterate on the software while Willison mostly chatted with it about what was happening. The approach echoes other efforts to let coding agents sense the physical world, such as using a webcam to debug hardware.

The outcome was a working tool that pinpointed the beep to within about 40cm, leading to the discovery of an old carbon monoxide detector. The project illustrates how AI coding agents can be extended beyond screen-based tasks to interact with and interpret real-world signals, potentially opening up new ways to solve physical-world problems with software.