“PMs who use AI will replace those who don’t”: Google’s AI product lead on the new PM toolkit | Marily Nika

Marily demonstrates using Perplexity's 'discussions and opinions' filter to mine Reddit for user sentiment on smart fridges. She creates two AI agents (pro and anti) that debate 20 rounds to identify the minimum feature set needed for product-market fit, completing comprehensive user research in under 3 minutes.

• •Use Perplexity's 'discussions and opinions' filter to search only Reddit threads for real user opinions
• •Prompt AI to create opposing personas (pro/anti) and have them debate to surface critical objections
• •Request output in specific format (features list) that becomes input for next tool in workflow
• •Extract minimum viable feature set from debate conclusions to ensure product-market fit
• •Complete market research that traditionally takes weeks in under 3 minutes

Using a custom GPT trained on her preferred PRD template, Marily pastes Perplexity's feature output and generates a complete product requirements document in 90 seconds. The custom GPT outputs PRDs in her voice with problem statements, user personas, feature prioritization, and architecture decisions.

• •Create custom GPTs with your preferred PRD template and writing style for consistent outputs
• •Paste feature lists from user research directly into PRD generator for seamless workflow
• •Custom GPT automatically generates problem statements, user personas, and feature prioritization
• •PRD serves as high-quality input for prototyping tools, reducing iteration cycles
• •Spend strategic thinking time instead of document formatting and copying

Marily demonstrates pasting the entire PRD into v0 to generate an interactive smart fridge dashboard prototype. The prototype includes features like door status monitoring, temperature controls, and GPU-powered local processing, creating a tangible vision for product reviews and stakeholder communication.

• •Paste complete PRD into v0 rather than short prompts for higher quality prototypes
• •Use prototypes as influence tools in product reviews to communicate vision more effectively
• •Interactive prototypes allow stakeholders to touch, feel, and experience the product concept
• •v0 interprets PRD requirements and adds logical UI features beyond explicit specifications
• •Prototypes add credibility and reduce need for deck-based product review presentations

Marily shows how to generate promotional videos using Google's VEO (via Flow) and OpenAI's Sora with cameos. She creates multiple video variations showing couples using the smart fridge, demonstrating how AI video generation can enhance product storytelling and stakeholder presentations, despite some humorous AI artifacts.

• •Use VEO through Google's Flow app to generate promotional videos from PRD features
• •Create Sora cameos with custom instructions (e.g., 'always make me look good') for personalized videos
• •Generate multiple video variations quickly to find the best stakeholder communication asset
• •Focus prompts on user narratives rather than feature lists for better video quality
• •Pair prototypes with AI-generated videos for compelling product review packages

The episode synthesizes the complete workflow: user research (Perplexity), PRD generation (ChatGPT), prototyping (v0), and video creation (VEO/Sora) - all completed in 15-20 minutes. This represents a fundamental shift in product management where AI-enhanced PMs will replace those who don't adapt to these new tools and workflows.

• •Complete end-to-end product development from research to stakeholder materials in 15-20 minutes
• •Tool hopping between specialized AI apps creates more powerful workflows than single tools
• •PMs who use AI will replace PMs who don't - it's a capability multiplier, not a replacement
• •Each tool's output becomes optimized input for the next tool in the chain
• •Focus shifts from document creation to strategic thinking and vision communication

Marily shares her creative use of NotebookLM to judge her AI Product Academy bootcamp demo day. She uploads all pitch recordings, provides judging criteria (innovation, impact, storytelling), and generates an interactive podcast that announces winners. The AI judges align closely with human judges and create an engaging experience for 200+ participants.

• •Upload multiple audio files (pitch recordings) to NotebookLM for batch analysis
• •Provide specific judging criteria in custom instructions for audio overview generation
• •Use interactive mode to have real-time Q&A with AI podcast hosts about results
• •Apply this technique to sales pitch competitions, hackathons, and training evaluations
• •NotebookLM's multi-file audio input/output creates engaging third-party analysis

When AI tools produce poor results (like screens inside refrigerators), Marily's strategy is to kill the instance, use AI to help write better prompts, and make prompts longer and more detailed. She emphasizes using AI to improve AI prompts rather than manually iterating.

• •Don't iterate on bad AI outputs - kill the instance and start fresh
• •Use AI (like ChatGPT) to help write better prompts for other AI tools
• •Longer, more detailed prompts produce better results with fewer iterations
• •Meta-prompting (AI helping write AI prompts) is more effective than manual prompt engineering
• •Apply this technique across all AI tools when results don't meet expectations