I've found it depends on the context (pardon the pun)
For example, personal projects that are small and where copilot has access to all the context it needs to make a suggestion - such as a script or small game - it has been really useful.
But in a real world large project for my day job, where it would need access to almost the entire code base to make any kind of useful suggestion that could help me build a feature, it's useless! And I'd argue this is when I need it.
It can ingest the entire codebase (up to its context length), but for some reason, I’ve always had much higher quality chats with smaller bite-sized pieces of code.
Autocomplete distracts me enough that it really needs to be close to 100% correct before it's useful. Otherwise it's just wrecking my flow and slowing me down.
Exponentially? Absolutely not. In the best case it creates something that’s almost useful. Are you working on large actual codebases or talking about some one off toy apps?
I spend most of my time thinking about what I'm trying to do and how to best achieve it, so code completion can only make me marginally more productive. If the tool can guess a large chunk of what I've decided to do, sure, that's nice, but at the end of the day it still only adds up to a couple minutes at best.
For example, personal projects that are small and where copilot has access to all the context it needs to make a suggestion - such as a script or small game - it has been really useful.
But in a real world large project for my day job, where it would need access to almost the entire code base to make any kind of useful suggestion that could help me build a feature, it's useless! And I'd argue this is when I need it.