Teaching an AI Agent to Talk to a Light Bulb (and Why I Had to Get Up From My Desk)

Short summary

An engineer used Claude to debug Sengled smart bulb pairing failures, discovering three stacked TLS issues (missing ALPN, cipher/version mismatch, stale certs) and implementing a fix that auto-regenerates certificates and forces compatible ciphers. The key insight: AI agents excel at code analysis but require humans for physical verification, creating a productive loop where 80% of debugging time is human testing and log collection.

• •AI agent analyzed TLS handshake failures and proposed cipher/ALPN configurations; human verified with hardware resets
• •Root causes were three nested TLS incompatibilities between ESP8266 mbedtls and modern OpenSSL defaults
• •Practical networking trick: dual NICs enable simultaneous connections to isolated networks during hardware pairing flows

Generated with AI, which can make mistakes.