I remember a few months ago, a top product manager in Silicon Valley mentioned that the ratio of engineers to PMs should gradually approach 1:2. Now it seems less reasonable; AI-driven efficiency improvements actually benefit product managers more because engineering implementation requires accuracy, whereas PRDs can be relatively more flexible in quality.

Sharing some practical experience from the past two days on how to quickly produce a PRD for a medium-complexity feature from scratch:

1. After determining priorities and feature scope, directly give a few natural language sentences to assistants like Manus/GPT/Gemini, requesting a comprehensive PRD, and compare the results (about 10 minutes).
2. Once you get the result (usually not perfect), trim some features based on your understanding to create a lean MVP, add considerations for boundary cases, and reference top competitors (about 30-60 minutes).
3. If there's no prototype, just upload it to Lovable; if it's an iterative requirement, take screenshots of the UI and feed the optimized prompt to Alloy. Remember to specify that the style must strictly follow the existing one, and think in advance about how many pages are needed and what they are (about 15-20 minutes).
4. After obtaining an interactive prototype, adjust it according to your aesthetic preferences (about 30 minutes).
5. Send the final prototype (screenshots or code) back to the agent that helped you create the PRD, asking it to revise the PRD accordingly (about 10 minutes).
6. Combine two relatively satisfactory versions into a final document and hand it over to designers and developers for evaluation and further refinement.

Recently, I interviewed many senior PMs from big companies. Hearing their lengthy product processes almost made me roll my eyes. The era belongs to teams with good intuition, quick understanding of requirements, rapid production, and fast validation.