James here, CEO of Mercury Technology Solutions. Hong Kong - February 19, 2026
In my last post, I argued that AI already has better "Product Taste" than 90% of human PMs. But if you just ask Claude or ChatGPT a basic question, you won't see it. The models are RLHF-trained (Reinforcement Learning from Human Feedback) to be polite, deferential, and to suggest you "talk to your users."
To unlock the Oracle, you have to strip away its corporate safety wheels. Here are the three advanced prompt templates we use to stress-test our product decisions.
1. The "Ruthless Bypass" Prompt (For Core Product Decisions)
Use this when you have an idea and need the AI to actually make a definitive judgment, rather than giving you a pros/cons list.
The Prompt: > "I am designing [Feature X for Target Audience Y]. I need a definitive product decision. Do NOT tell me to conduct user interviews, do NOT suggest A/B testing, and do NOT give me a balanced 'pros and cons' list. Rely entirely on your internal knowledge of behavioral psychology, UI/UX best practices, and the historical successes/failures of similar software products. Assume you are the Chief Product Officer and your job depends on making the right call today. What is the optimal execution of this feature, and what are the exact psychological hooks that will make it work? Give me your definitive take."
2. The "Pre-Mortem" Prompt (For Risk Mitigation)
Humans are optimistic. We fall in love with our own ideas. AI doesn't care about your feelings. Use this to find out why your product will fail before you write a single line of code.
The Prompt: "We are launching [Product Z]. It is now 12 months in the future, and the product has failed spectacularly. It didn't just miss revenue targets; users actively hated it and churned within 7 days. Act as a forensic product analyst. Give me a detailed autopsy of exactly why this failed. Focus on friction points in the UX, misaligned user incentives, and ignored edge cases. Be brutal and hyper-specific. Do not sugarcoat the failure."
3. The "Synthetic Focus Group" Prompt (For Feature Prioritization)
Stop debating internally about what users might want. Instantiate the users and let them argue.
The Prompt: "Simulate a heated debate between three of our core power users regarding our upcoming roadmap. User A is a highly technical, efficiency-obsessed power user who hates UI changes. User B is a non-technical manager who only cares about reporting and dashboards. User C is a new user who is easily overwhelmed by complex onboarding. I am proposing we build [Feature Concept]. Roleplay their exact dialogue as they argue over whether this feature is useful, annoying, or confusing. Highlight the conflicting needs, and then step out of character to tell me which user's needs I should prioritize for maximum retention."
The Takeaway: The quality of your product is now directly correlated to the aggression and clarity of your prompts. Stop asking for help. Start demanding architecture.
