Andreas Spiess

AI-Driven ESP32 Workflow (Spec → Code → Test) Using Claude Code

The video showcases an end‑to‑end AI‑driven workflow for ESP32 development, using Claude Code to generate, compile, flash, and test firmware for an iOS voice‑keyboard project that turns a smartphone’s speech‑to‑text capability into a USB keyboard via BLE. Key insights include a systematic eight‑step process—repository creation, idea description, functional specification generation, implementation planning, code generation, flashing, testing, and enhancement—automated by an AI agent. A custom "ESP32 workbench" built on a Raspberry Pi Zero 2W provides AI‑controlled serial access, Wi‑Fi access‑point creation, MQTT toggling, and real‑time log monitoring, solving both serial‑port scarcity and realistic testing constraints. Notable examples: the AI writes a complete functional specification from a markdown idea document, flashes the firmware through the workbench, and validates behavior such as Wi‑Fi dropouts and captive‑portal handling. The entire iOS keyboard prototype, covering BLE communication and USB HID emulation, was built and verified in roughly 1.5 hours, with the AI handling dozens of code iterations and test cycles. Implications are profound: embedded developers can now prototype complex IoT devices in hours rather than days, reduce reliance on manual debugging, and safely simulate network failures without impacting production environments. This workflow democratizes hardware development, making AI‑augmented engineering accessible to small teams and hobbyists alike.