When trying to code something that takes focus, Tab’s suggestions are quite distracting; there’s a chance that it’s going to get it right, so you check what it’s saying, but when it’s making fresh predictions multiple times per line, most of them wrong.. it isn’t great for flow. There’s a few ways to improve it imo
Tab completion delay option
If the user’s WPM is low and they keep on refusing suggestions, use that as signal to tune the model’s policy for suggestions
Have an option that is just an indicator that a completion is ready and can be peeked asynchronously to the user’s typing (eg a green dot in the gutter or something)
Let the user prompt the tab model itself, so that it could be asked to eg stop suggesting so frequently, or ways in which the suggestion could be improved
I typically snooze it. I’m finding that my default approach is snoozing it to start out with, most of the time.
We’ve noted your ideas about delay, a readiness indicator, adapting to WPM/declines, and prompting the model. Please add your vote in the threads above - it helps us prioritize.