The problem
More and more I use the Agent to get things done, and it’s awesome. But I still need to get in and code myself along the way. At this point, I kind of dread coding myself because for as amazing as the Agent is, the tab AI can be rather frustrating.
As an example, I just wrote something like this:
$domain_mode = new DomainMode($data['domain_mode']);
Great! The AI helped autocompleted this, and I was happy. A moment later it recommended this:
$domain_mode = DomainMode::cf_mode();
So it went from suggesting setting the domain mode based off the local data (correct) to just statically setting it to some arbitrary mode. 
In my experience the AI for tabbing is a bit overzealous. It’s constantly recommending things, even when I’m just sitting there, and will even undo something it just recommended!
The Idea
It would be great if the AI had a confidence value that it returned as part of the suggestion — confidence in how sure it is that this suggestion is truly valuable, or if it’s just shooting in the dark.
It would then be great to have a setting for determining the threshold for what actually shows up. Let’s say it’s a 0-1 percentage with a (0.8, 80%) default threshold. So anything below 80% confidence doesn’t show up. And I, as the user, can then increase or decrease the threshold based on how many suggestions I do or don’t want.
Thanks!