This is why I like GitHub's Copilot more than agents, it doesn't "do everything" it just incrementally completes what I'm working on, resembling how I write code, outputting just the next 5-20 lines that I can factor into my mental model of the software. It's slower than having AI do everything but it's much faster than writing it all myself, but it's probably not even a good trade-off today and if it is it certainly won't be for long.

The tools were sold on speed and quality, and they deliver both.

I think "quality" side depends on LOT of tuning you would need to keep in place. Quality might be good if you are dealing with single project, when you hare more than one repo to deal with, things need real attention and observed that quality goes down significantly.