Developer builds e-ink fridge magnet shopping list using Claude Code

By Splitfeed EditorialLast updated Sep 28, 2026, 13:10 UTC

On September 28, 2026, a developer posted to Hacker News a working shopping list application built on the M5Stack PaperMono, a compact device combining an ESP32-S3 chip with an e-ink touchscreen. The firmware, around 2,400 lines of C++, handles Wi-Fi synchronization between the device and a mobile web app, and continues functioning offline when a connection is unavailable. The developer noted that the PaperMono board also supports BLE and LoRa, though neither is used in this particular build.

The detail drawing the most attention is that the developer wrote none of the C++ by hand. The entire codebase was generated by Claude Code, and the developer framed the project explicitly as a test of how well the tool handles development for hardware it would not have been trained on extensively. That framing matters because embedded C++ for specific microcontroller boards is a narrower and less forgiving domain than general web development: memory constraints, peripheral drivers and real-time behavior leave less room for generic or hallucinated code to pass unnoticed. A 2,400-line firmware that runs on physical hardware and syncs reliably over Wi-Fi represents a more concrete benchmark than a browser demo.

None of the three camps have published reactions to this story yet, but their typical positions are easy to anticipate. The Pro-AI camp would likely treat this as evidence that AI coding tools are now capable enough to handle specialist, hardware-level work with no manual intervention, lowering the barrier for hobbyists and non-embedded developers. The Anti-AI camp would probably question how the developer validated the generated code, raise concerns about whether vibe-coded firmware introduces subtle bugs that only surface under edge conditions, and push back on framing generated output as a reliable measure of a tool's capability. The Middle Ground camp would likely acknowledge the practical result while arguing that the real test is maintainability, asking how easily someone could debug or extend 2,400 lines of C++ they did not write and do not fully understand.

The argument would be sharpened if the developer open-sources the full codebase, which would allow others to audit the generated C++ for correctness and idiomatic quality. Any follow-up post describing bugs found after publication, or extensions built on top of the existing firmware, would also give a clearer picture of how well AI-generated embedded code holds up beyond the initial working demo.