Total BS. You don't show the robots a prototype model, then they look at it with their robot eyes and learn to make it with "AI", duplicating the errors in your prototype. You design it with CAD programs, then software can use those models, or you manually train the robots with a pendant. Rossi may have read somewhere about research into this, thought it sounded cool and knows he can fool his cult who don't understand what the buzzword AI really means. Sure some other BS company may claim something like this, but it doesn't mean it works. The whole field of AI has basically overpromised and under delivered, and I have basicically working in this field for 34 years. We called it "AI" back in the 80's, and that buzzword has had yet another resurgence. But now it is "Deep" learning. To match Rossi's "Deep BS".
Rossi is *not* saying that AI will look at a prototype and figure out how to make it, far from it. As I understand, he's only saying that humans make prototypes, and these rough hewn prototypes have to be tested to make sure they work as intended, and only then can robotized production be programmed and implemented. Otherwise robots - which generally only do what they're told - would be making potential duds, for the very reason they're not smart enough (generally) to know what works or not.
Speaking of AI, I find it very interesting that there is allegedly an AI component to the functioning of the Ecat, which was also stepped up with Rossi's recent SKL enhanced upgrade. If it is described in more detail in the future it looks like you will be in a good position to critique it.