AI Audio: Less Alts, More Metadata, or Both?

Hey everyone, Diana Ruiz-Perez here. I'm an AD/Producer, currently prepping a short film shot on a C70 with some Sound Devices 833s on set for audio. I've been experimenting with some of these new AI audio cleanup tools (the 'remove background noise' and 'separate dialogue' features are genuinely impressive in preliminary tests on practice footage. They're making me rethink some of my usual production audio workflows. For example, traditionally, I'm a stickler for getting a wide range of 'alts') room tones, wild lines, clean effects, to give post-production every possible option. With these AI tools seemingly able to isolate and clean up so much, it makes me wonder if the emphasis on capturing every single 'alt' is becoming less critical. Or, conversely, are these tools simply raising the bar for what post expects, pushing us to be even more meticulous with metadata for AI processing, and requiring even more precise delivery packages? What's your take on this balance?