Is product discovery the last part of software development that AI hasn't touched yet?

Since Last Tuesday night I have been thinking of this problem and it really has hit me hard. Engineers have Cursor. Designers have AI-assisted Figma plugins. But PMs are still manually reading through 50 tickets, switching between Amplitude and Notion, and writing PRDs by hand. I've been thinking about why AI hasn't meaningfully improved the discovery side of product development - the "what should we build" question - even as it's dramatically accelerated the "how do we build it" side. Curious what this community thinks.