A 1994 Jeep Cherokee isn’t exactly the first vehicle you’d associate with artificial intelligence. Its engine computer predates modern OBD-II diagnostics, making something as simple as monitoring live engine data surprisingly difficult.
That didn’t stop Muhammad Zulali from connecting a $3 Arduino Nano to his Jeep and asking Claude AI to investigate. What began as an attempt to diagnose an emissions problem eventually became a much more ambitious experiment.
As Zulali demonstrates on his YouTube channel, Claude Code and OpenAI’s Codex helped him communicate with the Cherokee’s original engine computer. Within roughly a week, the project had progressed from reading sensor values to executing custom code inside the aging electronics.
The motivation was practical: his Jeep kept failing California’s emissions inspection, and repeated testing was getting expensive. Finding the cause would require understanding a computer that was never designed for today’s diagnostic tools.
A $3 Arduino Opens the Door
The Cherokee was failing its emissions test because of excessive nitrogen oxides, or NOx, particularly at 25 mph. Replacing the oxygen sensor hadn’t solved the problem, and each additional inspection cost approximately $100.
Claude identified Chrysler’s older SCI communication protocol and helped Zulali connect an Arduino Nano to the diagnostic wiring. After some troubleshooting, the setup began reporting coolant temperature, battery voltage, and other engine data.
From Reading Sensors to Controlling Hardware
The breakthrough came when Claude successfully commanded the Jeep’s fuel pump to run, rather than merely observing sensor readings. Zulali then logged driving data and investigated confusing fuel-trim figures, eventually discovering that an existing diagnostic tool was interpreting them incorrectly.
Further analysis of the engine computer’s copied program revealed differences between California and federal fuel calibration tables. Zulali suspected those differences might contribute to his emissions failure, although the video doesn’t establish them as the sole cause.
Running New Code on a 1994 Computer
Reprogramming the factory computer proved impossible because its hardware predated a later programmable version. Codex found a way to execute custom instructions through the existing diagnostic connection, bypassing the need for a conventional ECU reflash.
The experiment progressed to writing code into the computer’s persistent memory, allowing the modifications to survive a power interruption. After extensive testing and another frustrating trip to the emissions station, Zulali reported that the Cherokee finally passed.
The achievement wasn’t autonomous driving or AI taking over the steering wheel. It was something that’s arguably more interesting for enthusiasts: modern coding agents gaining diagnostic and limited control capabilities over a three-decade-old engine computer, using inexpensive hardware.