The other 75% is the stuff you actually have to think about.
This feels like saying linters impact x0% of code. This just feels like an extension of that.
It's just a very advanced autocomplete, completely integrated into the internal codebase and IDE. You can read this on the research blog (maybe if everyone just read the blog).
e.g.
I start typing `var notificationManager`
It would suggest `= (Notification Manager) context.getSystemService(NOTIFICATION_MANAGER);`
If you've done Android then you know how much boilerplate there is to suggest.
I press Ctrl+Enter or something to accept the suggestion.
Voila, more than 50% of that code was written by AI.
> blindly committing AI code
Even before AI, no one blindly accepts autocomplete.
A lot of headline-readers seem to imagine some sort of semi-autonomous or prompt based code generation that writes whole blocks of code to then be blindly accepted by engineers.
I’m pretty sure the actual ratio is much lower than that. In other words, LLMs are currently not good enough to remove the majority of chores, even with the state of the art model trained on highly curated dataset.