After his mother passed away, the author inherited a box of old film negatives and bought a Plustek 8200i scanner, only to find its chipset had changed to one unsupported under Linux. Earlier, he had used Claude Code to build Linux configuration tools for his Dell camera controller, so he asked whether Claude could help him write a working SANE driver. He got stuck and shelved the project for months.
Then he saw a post by Tom Crider, aka Glorious Eggroll, describing how he had used AI tooling to fix a mouse driver, with an agent controlling a Windows instance to capture the data needed. The author applied the same approach: he ran the vendor's Windows driver in a KVM virtual machine with the scanner passed through, captured USB traffic on the Linux host with usbmon, and let Claude compare the traces. Within a few hours he had a first working scan, and days later a near-perfect driver with infrared support. By then, however, someone had already committed scanner support to SANE. He also wrote a negative-scanning app he called Negative Vibes.
Encouraged, he had Claude reuse the harness to fix the MIPI camera on his laptop, and by the end of the day it had patches for both the kernel and libcamera. He then asked Claude to make the harness generic and documented, so others can try it on their own hardware.
The approach suits the tedious work of reverse engineering undocumented hardware. An AI model can scan thousands of lines of USB capture, spot repeating sequences, propose what changing bytes mean, and write tests to confirm them. The human still decides what to build and verifies the result on real hardware, but the effort shrinks considerably.
Source: Hacker News · Summarized by HeadlinesBriefing