AI ESP32 Coding without Losing Control
on
AI ESP32 coding can turn a plain-English description and a few hardware details into a working Arduino sketch in seconds. In this Elektor TV clip from the ESP32 by Example episode of Elektor Engineering Insights, author and Villanova University adjunct professor Jim Solderitsch explains how he uses that speed as a learning accelerator while keeping the learner responsible for understanding the code.
AI ESP32 Coding as a Learning Accelerator
Solderitsch starts with a deliberately bounded task, meaning he’s not asking an AI system to invent an entire product. For the ESP32 by Example Kit (EEK), he describes the example he wants in ordinary English and supplies context files that define the hardware, including GPIO assignments and device information. The model then produces an example that can be inspected, loaded into the Arduino IDE, compiled, and tested on the board.
That changes the amount of time needed to create teaching examples. Solderitsch says the curated examples behind his course took years of investigation, prototyping, and testing. Once the basic hardware and programming concepts are understood, generating another small example can otherwise become repetitive work: assembling code, checking libraries, and adapting it to the target board.
Give the Model Hardware Context
AI ESP32 coding becomes more useful when the model knows exactly what hardware it is supposed to target. Solderitsch gives the AI a description of the EEK plus structured information about its connections, then asks for a specific behavior. His test cases include examples for LEDs, switches, the OLED, and the motion sensor.
The generated sketch is still something to read and understand. Solderitsch sums up his approach simply: “I want AI to teach me how to write better software.” In his university teaching, the AI-generated examples sit alongside the Arduino IDE and the physical hardware, so students can see what the model produced and what the ESP32 actually does with it.
Generated Code Still Needs Testing
The method also gives AI failures a useful place in the lesson. Solderitsch says some generated examples contained trivial mistakes. Those errors can be compiled, tested, identified, and corrected rather than hidden behind an apparently successful prompt. A plausible-looking sketch is only a starting point until it behaves correctly on the real hardware.
During the wider discussion, Solderitsch also talks about Claude, Google Antigravity, Ollama, M5Stack hardware, simulators, and drone control. You can watch the full EEI #62 discussion on Elektor TV.
Discussion (0 comments)