PMs seek 'Cursor' for product discovery

The "what to build" problem is fundamentally harder for AI to solve than "how to build it" because it requires synthesizing messy human signals, not just predicting tokens.

• –Fragmented data across support tickets, Amplitude, and Notion creates a context gap that general-purpose LLMs struggle to bridge without deep ecosystem integration.
• –Predictive models naturally favor execution (how) over the novel synthesis required for breakthrough discovery (what).
• –ChatPRD and similar tools are moving toward "agentic discovery," where AI analyzes customer signals to suggest features rather than just drafting the PRD for a pre-determined idea.
• –Sceptics argue that interpreting human nuance is a "last mile" problem where AI often misses the strategic "why" that defines great product leadership.