Trying Pair Programming With An LLM Chatbot

An experienced embedded developer with a self-described aversion to pair programming tests GitHub Copilot as a coding assistant on two tasks: STM32/CMSIS C++ embedded code and Ada networking. The experiment quickly goes sideways when Copilot ignores the CMSIS requirement and generates non-functional code with incorrect macros and missing headers. The author concludes that Copilot is unreliable for specialized embedded work, produces output requiring extensive error-checking, and agrees with research suggesting LLM chatbots degrade critical thinking. The verdict: back to documentation and search engines.