Yes, visual recognition by AI is a starting point, but it is not the key link for a robotic hand to hold an apple without turning it into puree.
What really matters is the hand’s ability to feel pressure and adjust its grip, regardless of how good the AI is at identifying the fruit.
If the hand sensors don't work well, or if the mechanics are too rigid, no matter how loudly the AI shouts "apple!", the result will be the same: a disaster.
Imagine you can read a perfect sheet music, but your instrument is broken; the music will never come out correctly.