How I AI“PMs who use AI will replace those who don’t”: Google’s AI product lead on the new PM toolkit
CHAPTERS
- 0:00 – 2:54
Why “AI-enhanced PM” is the new baseline (meet Marily Nika)
Marily Nika introduces her background as an AI PM and frames the shift from being a PM for AI products to being a PM who uses AI in everyday work. Claire sets up the episode as a rapid, tool-driven walkthrough of a modern PM workflow—from discovery through stakeholder influence.
- •Marily’s transition from “pure AI PM” (model training era) to “AI-enhanced PM”
- •Goal: more impact, productivity, and sharper product thinking using new tools
- •Episode promise: speed-run research → PRD → prototype → video to sell the vision
- •Positioning: AI changes the craft and cadence of product management
- 2:54 – 6:15
The smart-fridge moment: a viral insight becomes a product wedge
A surprising fridge notification (“items expiring soon”) becomes the spark for a full product exploration. The hosts use it as a concrete, consumer-focused example to demonstrate AI-powered PM techniques end-to-end.
- •Real-world trigger: fridge detects “expired” items (e.g., 80-day-old Coke)
- •LinkedIn post goes viral; PMs brainstorm possibilities in comments
- •Initial critique: current UX (e.g., ‘search recipes’) feels weak/underdeveloped
- •Use case framing: smart fridge for busy families and household chaos
- 6:15 – 11:19
Ultra-fast user research with Perplexity: mining Reddit for real opinions
Marily shows how Perplexity’s Reddit/discussions mode can replace days of early discovery with minutes of synthesized sentiment and threads. The goal is to extract adoption drivers, objections, and patterns directly from authentic user conversations.
- •Perplexity filter: search “discussions/opinions” to mine Reddit at scale
- •Instant access to positive use cases, concerns, and adoption barriers
- •Ability to jump into cited threads for source context and nuance
- •Outcome: quick, evidence-backed perspective to guide feature choices
- 11:19 – 13:40
From feature list to PRD in ~90 seconds: a custom ChatGPT PRD generator
Marily copies the debated feature set into a purpose-built “AI Product GPT” that outputs a structured PRD in her preferred template and voice. The PRD becomes a high-quality artifact for alignment and downstream prototyping.
- •“Tool hopping” workflow: Perplexity outputs become ChatGPT inputs
- •Custom GPT includes a PRD template and style matching Marily’s voice
- •PRD elements: problem statement, target users, privacy constraints, architecture, prioritization
- •Benefit: massive time savings; PM time shifts from formatting to strategy
- 13:40 – 16:20
Interactive prototype with v0: turning the PRD into a clickable smart-fridge UI
Marily pastes the full PRD into v0 and generates a dashboard-style interface for a smart fridge. The resulting UI includes both requested requirements and “guessed” helpful additions, making the concept tangible for stakeholders.
- •Prompt: “Create the UI of a smart fridge, given this PRD”
- •Generated UI elements: temp controls, door status, power usage, recent activity, safety monitoring
- •Tool adds plausible features beyond the PRD (useful but needs PM judgment)
- •Notable theme: local processing/privacy surfaced visually (e.g., badges/diagnostics)
- 16:20 – 21:30
Prototypes as influence: winning product reviews with something people can feel
The prototype is positioned as a persuasive artifact for product reviews and cross-functional alignment. Instead of debating decks and docs, stakeholders can click, explore, and internalize the vision quickly—boosting credibility and decision speed.
- •Best usage moment: product reviews where investment decisions happen
- •Interactive prototypes outperform PRDs-turned-decks for vision communication
- •Tactile experience increases buy-in and reduces ambiguity in discussion
- •AI accelerates review cadence; prototypes keep up with faster cycles
- 21:30 – 30:17
From prototype to pitch: generating promo videos with Flow (Veo) and Sora
Marily extends the workflow into video generation to help stakeholders “see” the user story. They test Google Labs Flow/Veo and compare with OpenAI Sora, noting impressive quality alongside common generative quirks.
- •Flow/Veo setup: paste PRD/features → choose quality, aspect ratio, audio settings
- •Results show typical issues (odd screens inside fridge, over-indexing on door-ajar)
- •Claire’s advice: use a user narrative/hero story rather than a feature dump
- •Sora cameo workflow: create avatar + custom instructions; generate a skit (even with Mark Cuban)
- 30:17 – 32:06
The 15–20 minute end-to-end PM workflow (research → PRD → prototype → video)
Claire recaps the full speed-run as a new “PM package” that can be produced in under 20 minutes: defensible insights, a structured PRD, a clickable prototype, and compelling visuals. Marily underscores the competitive shift: PMs who use AI outpace those who don’t.
- •End-to-end deliverables: Reddit research, feature set, PRD, prototype, promo video
- •New leverage: faster persuasion and clearer storytelling for internal stakeholders
- •Reframing: AI isn’t replacing PMs—AI-using PMs replace non-users
- •Workflow is customizable; tool choices depend on the individual PM’s style
- 32:06 – 37:38
NotebookLM as an AI judge: scoring demo days with multi-audio analysis + hosts
Marily shares a novel education/workflow use case: using NotebookLM to evaluate multiple startup/demo pitches by ingesting audio files and generating rankings and an announcer-style audio overview. The interactive “call-in” mode adds a fun, engaging layer for live events.
- •Use case: AI Product Academy demo day—record each pitch and upload files
- •Custom judging criteria: innovation, impact, storytelling; pick top three
- •NotebookLM outputs an audio overview and supports interactive Q&A mode
- •Broader applications: hack weeks, pitch competitions, sales enablement roleplays
- 37:38
When tools fail: reset, prompt with AI, and be radically specific
In the lightning round, Marily explains how she handles poor generations (e.g., weird video artifacts). Her tactic is to stop wrestling with a broken thread, restart clean, and use AI to help write a longer, more precise prompt to reduce iterations.
- •Best practice: “kill the instance” and start over when stuck
- •Use AI to write better prompts (meta-prompting)
- •Longer, more detailed prompts reduce rework and ambiguity
- •General principle: treat prompting like spec-writing—clarity wins
Critical thinking hack: pro/contra agents debate to surface minimum viable features
To counter LLM agreeableness, Marily prompts Perplexity to create two opposing agents and run a multi-round debate. The debate is then distilled into the minimum feature set needed to convince skeptics—an explicit path toward product-market fit.
- •Prompt pattern: one agent pro smart-fridge, one agent against it
- •Run ~20 rounds of debate grounded in referenced discussions
- •Extract “minimum set of features” to convert skeptics
- •Claire’s takeaway: adversarial personas improve rigor and reduce bias
Why better inputs win: PRDs as the scaffold for higher-quality prototyping
Before prototyping, the hosts discuss experimentation and why jumping straight to UI prompts often increases rework. Investing in a solid PRD reduces iteration fatigue and produces more coherent prototypes.
- •Marily’s lesson: moving fast too early creates exhausting downstream tweaks
- •PRD provides structure + context vs. isolated UI prompts
- •Combining content constraints (privacy/local-first) with UX needs improves outputs
- •Prototyping is framed as communication and clarity, not just design
