Somewhere among unpacked moving boxes, something was beeping, just too rarely to find by ear. Developer and AI blogger Simon Willison had Claude Code build a program that found the beep to within 40 centimetres. The interesting part is not the beep, but how he went about it.
What happened
Five ordinary devices became one listening network
On 9 November 2026 he described Beep Locator: a Python web server on his laptop that shows a floor plan. Other devices join as listeners through a QR code or link. He used five: two iPads, a phone, a MacBook Pro and a Surface Ultra.
A rough mode rescued the failed first idea
The devices were meant to place themselves on the map by playing sounds to each other. According to Willison, that did not work well. In "Rough mode" he drags the devices roughly into place himself and the system refines that with the sounds.
Clapping first tested whether the calculation was right
He clapped in different spots and so saw whether the calculated position was right, before he trusted the real beep.
Willison decided himself which sound was the beep
- He could play back every captured sound and mark it as "the" beep.
- The system then listened for it specifically and pointed to a spot, to within about 40 centimetres.
- In the boxes he found an old carbon monoxide alarm asking for new batteries.
The first instruction was a problem description, not software
He started in Claude Code in the cloud, from his iPhone, with Opus 5.5. His first instruction: write down in a file what we want to solve. Along the way Claude found a 2007 paper on measuring distances with sound between ordinary devices. On his laptop he carried on building with the newly released Fable 5.2. He mostly watched, asked for explanations and occasionally for a feature.
What this changes for you
Nobody in an SME is hunting for a beeping alarm, but almost every business has jobs that no package covers and nobody has time for. An AI agent can build its own tool for that in a short time. That does ask something of you.
As a managing director or operations manager:
- Start with the problem. Write down what goes wrong, what the outcome should be and what you have already tried.
- Test with an answer you already know. Run the tool on a case whose outcome you know first, as Willison did with his claps.
- Keep the decision with a person. Agree in advance where the agent decides and where a person signs off.
As the person responsible for IT, you set the boundaries. Here the server could send new code to every device and microphones were listening all the time. At home that is a handy experiment; at the office you want to know which devices take part, who sends code to them, what gets recorded and where it ends up. A throwaway tool should not slip into production unnoticed.
Our view: first the problem and the desired outcome on paper, then the tool, because only then can you test whether it does what it should. The same goes for an Odoo implementation.
How it works under the hood
The devices talk to the server over WebSockets and listen through the browser. Each keeps 60 seconds of audio and searches for beeps with a Fourier transform and a high-pass filter. When a beep is heard, the device with the fastest connection plays a short chirp as a reference point; the position follows from the clips of all listeners. This builds on BeepBeep, the 2007 paper by Chunyi Peng and colleagues. With installModule() the server sent new code to the listeners, so they did not ask for the microphone again after every reload. That also lets it run arbitrary code on every device. Claude hears nothing itself, but runs code that analyses recordings: Willison calls it giving Claude Code ears.
This news touches chapters 14 and 23 of the course AI(l) or nothing by Newway. If you follow the course, this subject returns there at a calmer pace.
What you can do with this
Want to know what this means for your business? Plan an introduction: half an hour, no sales pitch, but an honest answer to the question where to begin. Rather read on first? The course AI(l) or nothing is free and starts with the basics.
Written in response to a publication by Simon Willison: https://simonwillison.net/2026/Nov/9/mysterious-beeps/
S.O.S