I tested Copilot in an existing open-source codebase with unit tests [video]

He probably spent more time figuring out if the suggestions are usable than he would have needed to just implement those manually.

it would be interesting if copilot could generate the code based on your unit tests

Is there a TL;DR?