I find it interesting to divide it into “problems we think are solvable” and “problems that can’t be resolved to a single solution.” For example, the Grammerly mistake you mention is something that should, with a better AI, be improved relatively easily. Are Copilot’s problems the same thing, or are some or them tied up with ambiguous issues that it can’t solve without more contextual information that just isn’t in the code? Our tendency is to assume the latter, but occasionally AI gives us that feeling of surprise when it “guesses” something that we didn’t think it had enough information to solve.

So while I’m not exactly a Copilot fan, I’m still ready to be surprised (again)…