Well, Actually: AI Coding Is Best at Glue, Not Genius
The author fed two GitHub repositories into OpenAI's Codex and had it write a Python script to fix a smart home problem. Specifically, the task involved integrating a Yale smart lock with Home Assistant. The AI did not invent novel algorithms. It connected existing code that the user lacked the patience to wire together manually.
This teaches you to stop asking AI to solve hard problems from first principles. Instead, deploy it where you have two working pieces and need the duct tape. Your workflow shifts: collect repositories, describe the gap, and let the model handle the tedious integration you would otherwise procrastinate.
A How-To Geek writer conducted this experiment with Codex, OpenAI's coding model. The repositories were existing open-source projects for Yale lock integration and Home Assistant.
Step 1: Open ChatGPT, Claude, or another chatbot with code capabilities and paste two related GitHub repository URLs that solve adjacent pieces of a problem you have. Step 2: Write a plain-English prompt explaining what you want to connect between them, then ask the AI to write a script. Step 3: Run the script in a free online Python environment such as Replit or Google Colab, and verify whether the integration works as described.