An image of a word clock is used as a prompt for ChatGPT-4o to generate the initial HTML structure and CSS layout.
Natural language comments in TypeScript are used to guide GitHub Copilot's statement completion to generate a time conversion function.
A man-in-the-middle proxy reveals how Visual Studio Code sends code context to GitHub Copilot's backend for completion.
GitHub Copilot Chat is used to generate a time-rounding function, revealing its use of GPT-4 and a different system prompt than statement completion.
GitHub Copilot is prompted to write a comprehensive suite of unit tests using the Vitest framework for the previously generated functions.
The final step involves prompting the AI to generate the complex business logic that maps the current time to the correct words on the clock face.
The discussion covers Copilot's limited project context, token limits, the benefits of strictly-typed languages, and free alternatives.
30:24Markus Walker
29:35Liuba Gonta & Yuliya Khadasevich
58:06Anna Fritsch-Weninger
39:11Thomas Dohmke
47:35Laurie Voss, Rey Bango, Hannah Foxwell, Rizel Scarlett & Thomas Steiner
31:56Prashanth Chandrasekar, Alejandro Saucedo, Jakob von Lindern & Demetris Cheatham
26:52Thomas Dohmke & Demetris Cheatham
28:09Jan Oberhauser, Mirko Novakovic, Alex Laubscher & Keno Dreßel
Jobs that call for the skills explored in this talk.
