I like where Intercom is going with this, but I feel like ultimately it has one of the same faults as CSAT: it measures the conversation, not the end-to-end experience. Also, because it's all based in AI prediction and reasoning, you aren't getting any of the context that a customer hasn't shared in their conversation that's driving their sentiment (or even, perhaps, resolution status).
But, I think those are solvable problems to a degree and I like that Intercom is thinking about that, instead of Zendesk being over here saying "in the year of our lord 2025, you can do a Likert scale in CSAT now"