Muhammad Zulali used Claude, an AI model, to diagnose and fix emissions problems on his 1994 Jeep Cherokee XJ. The vehicle failed California emissions tests and lacked an OBD-II port, making standard diagnostic tools incompatible.

Zulali asked Claude to identify the Cherokee's onboard computer communication protocol. Claude suggested using a $3 Arduino Nano microcontroller to translate the signal and read data from the car's computer. The initial readings proved inconclusive, leading Zulali to test fuel pressure. Claude then controlled the fuel pump directly to raise pressure, though this didn't resolve the underlying issue due to faulty data.

Zulali had Claude copy the entire computer code for offline analysis and identified a fuel-mapping problem causing the emissions test failure. He then used Codex, OpenAI's coding agent, to reflash the computer because the hardware lacked conventional reprogramming capability. After the reflash, the Cherokee passed emissions testing.

Zulali completed the work without visiting a shop, though he acknowledged the process requires significant programming knowledge. He noted that most vehicle owners would likely find a mechanic more practical.