I would say the copilot system isn't really there yet for these kinds of changes, you don't have to run experiments on a language framework to figure that out.
The answer is probably that the Copilot team is using the rest of the engineering organization as testers. Great for the Copilot team, frustrating for everyone else.
For it to be "failed" it would have to also be finished/completed. They are likely continuously making tweaks, this thing was just released.
"It would have to be finished/completed"
Do you honestly not see a problem with those two statements in such close proximity? Is it finished or is it released? The former is supposed to be a prerequisite for the latter.
We can debate whether they should have called this an experiment or an alpha or beta or whatever, but that's a different discussion.
The fact that people are using it currently does not make it a failure. When MS shuts it down, or Copilot is wildly unprofitable for multiple quarters, team behind it quits, etc, etc, then we can determine whether it has failed or not.
But if they continue to have paying customers and users are finding some benefits over not having Copilot, and MS continues to improve it (doesn't let it rot), then you'd have to provide some evidence of its failure that isn't "look at Copilot being stupid sometimes". Especially when stupidity is expected of it.