By skyforbes 
I used ChatGPT to design Nixie‑style digits, then asked Manus — an AI coding tool — to generate ESP32‑S3 firmware for six 0.96‑inch ST7735 LCDs. Here’s what worked, what broke, and how I fixed it: from hardware SPI and SPIFFS image handling to timers and time zones.
Some time ago I bought a 0.96‑inch 80×160 LCD and wanted to build a Nixie‑style clock.
The very first step already involved AI: I used ChatGPT to generate digits with a Nixie‑tube look, sliced them into ten images (0–9), converted them to RGB565 arrays, and tested the display. The effect looked pretty good.
That got me thinking: could AI also generate the firmware for this clock? AI is already proven to be highly efficient for some front‑end work — could it offer the same productivity in embedded development, at least for a simple project?
As it happened, Manus had just opened registration, so I took it for a spin.
Requirements
To keep the scope tight and reduce the chance of failure, I defined a minimal feature set: use an ESP32 to drive six LCDs that display the sliced digit images, and provide APIs for image updates plus AP‑mode configuration.