Engineer Ben Zhang said he recovered a lost phone in his office after asking Claude to build a Bluetooth signal strength tracker, a workaround he used because Apple’s Find My had been disabled by MDM, or mobile device management.
Zhang wrote that he spent 30 minutes turning the office over without finding the device. With Find My unavailable under the company’s device controls, he asked Claude how he could locate the phone. The AI suggested tracking Bluetooth signal strength and wrote a meter for him in about a minute. He then walked around the office watching the numbers climb and eventually found the phone.
MDM removed the most direct location tool
Companies often use MDM systems to manage employee devices, including remote software installation, password policies, and lock or wipe functions for lost hardware. One trade-off is that Apple’s native Find My feature is often disabled at the same time.
For employers, that can be part of a security setup. For an employee who has misplaced a phone, it removes the most obvious recovery tool. That was the situation Zhang described.
How findphone works
Zhang later released the tool on GitHub under the name findphone. Its core function is to continuously read a Bluetooth device’s RSSI, short for received signal strength indicator. In practical terms, a stronger reading usually means the target device is closer.
The program is written in Swift and requires macOS 13 or later. It does not rely on external packages, and it can be compiled with swift build -c release. Zhang also included a script that packages a universal binary for both arm64 and x86_64 Macs.
Two extra options for usability
- With the sound flag, the tool emits a reverse-parking-style tone that speeds up as the user gets closer to the target.
- With the redact flag, it masks the Bluetooth address, which helps avoid exposing device information in screen recordings.
Useful for direction, not precise positioning
The report also noted the limits of Bluetooth-based positioning. RSSI typically offers accuracy in the 2-to-5-meter range, and even in favorable conditions it is hard to get within 2 meters. In spaces with metal, glass, or water, reflection and diffraction can push the error beyond 5 meters and make the readings less predictable.
That leaves the tool better suited to showing whether the user is getting warmer or colder than to providing an exact location. The final step still depends on the person walking the space and checking the surroundings.
A one-off problem turned into a working utility
The article frames the episode as a sign of falling software creation costs. In the past, building a custom tool for a single personal inconvenience often would not have been worth the effort. With AI, that calculation can change, making it possible to create narrowly targeted software quickly enough for it to be useful in the moment.