I tested Copilot in an existing open-source codebase with unit tests [video]
youtube.com
youtube.com
Seems to me that once you can go either way then you just proved the other redundant.
https://rbcs-us.com/documents/Why-Most-Unit-Testing-is-Waste...
^ where the 'unit' is basically a function. i.e., to test a function fully, write a test as complicated as the function
then creats a few unit tests.
This was an interesting watch and useful to see other peoples setups.
Presumably value "unlocked" for GitHub investors, by leveraging the ML model that drives some product based upon Copilot like a next-gen IDE. I agree it's not a good risk to assume for the users of Copilot.