Aakash GuptaI got a private Masterclass in AI PM from Google AI PM Director
CHAPTERS
- 1:38 – 3:03
Meet Jacqueline Kunzelmann (Google AI Product Director): AI PM roles are real (and how leveling works)
Jacqueline confirms AI PM is a real, distinct role focused on AI-native products. She and Aakash discuss compensation broadly and clarify how Google PM levels compare to startup titles, emphasizing calibration by scope and years of experience.
- •AI PM is a legitimate role focused on AI-based/AI-native products
- •Compensation context and why AI PM is a well-paid, experience-heavy track
- •How Google PM leveling maps (L3/L4 early career; L5+ with experience)
- •Why “Senior PM” differs dramatically between startups and Google
- 3:03 – 6:10
How AI changes product building: faster iteration, new product types, and embracing discomfort
Jacqueline explains AI’s dual impact: it changes how teams build (tools that accelerate shipping) and what products are worth building (new capabilities change feature possibilities). She also normalizes the overwhelm of rapid model progress and reframes discomfort as a sign you may be doing true 10x work.
- •Two shifts: AI-powered building process + AI-enabled product capabilities
- •The pace of model releases makes building both exciting and overwhelming
- •Zero-to-one remains messy; AI makes it ‘sparkly’ but still ambiguous
- •10x thinking should feel uncomfortable—don’t confuse discomfort with wrong
- 6:10 – 17:54
Demo 1: Colorizing old photos—prompt structure, iteration loop, and negative prompts
Using a wedding photo of her grandparents, Jacqueline demonstrates photo restoration and colorization with a carefully tuned prompt. She breaks down how she iterates by asking Gemini to refine prompts based on failed outputs, including adding negative prompts to avoid common artifacts.
- •How to craft a long-form prompt: color targets, lighting, texture, optics
- •Using Gemini to debug and improve prompts (“why didn’t this work?”)
- •When and why to use negative prompts to fix repeated issues
- •Takeaway: prompting is an iterative skill, not a one-shot input
- 17:54 – 20:17
Ads break + Demo 2: Pet → drone show image → Veo video (and consistency tips)
After sponsor messages, the episode returns to a chained workflow: generating a drone-show version of a pet image, then turning it into a video with Veo. Jacqueline shares prompt philosophy (“interpret, don’t copy”), notes how delight can emerge (tail wagging), and offers a method for scene consistency using seed images and tools like Flow.
- •Chaining: generate drone-show image, then animate with Veo video
- •Prompting principle: specify non-negotiables; ‘interpret, don’t copy’
- •Veo prompt simplicity can still work, but more instruction yields control
- •Scene consistency strategy: keep a consistent character via image iteration, then use as Veo seed
- •Mention of Flow for building longer, more consistent video sequences
- 20:17 – 28:41
Demo 3: Opal—building mini AI apps by chaining prompts and models
Jacqueline introduces Opal as a natural-language way to build, edit, and share “mini AI apps” by chaining model calls. She demos simple and advanced flows (nature collage, custom storybook) and shows how Opal can generate an app from a text instruction plus a URL, then expose the underlying prompts and model choices.
- •Opal’s value: prompt chains + user inputs + configurable outputs
- •Examples: photo → nature collage; storybook maker with consistent illustration style
- •Natural-language creation: describe app + paste a URL as rubric/criteria
- •Inspect and edit prompts, change models (e.g., Flash vs Pro), share/remix apps
- •Output targets: web UI now; expanding to Docs/Sheets and more integrations
- 28:41 – 33:04
When to use chat vs workflow apps vs full agents—and why ‘building in public’ matters
Jacqueline advises using tools like AI Studio and Opal to prototype quickly, validate ideas, and stress test feasibility before committing to production. She shares her “10 side projects” approach to staying creative, thinking bigger, and generating public feedback loops early.
- •Start with idea conviction; prototype before production
- •Chat is powerful for exploration; workflows help repeat/share; agents for autonomy
- •Building in public to get early signal and user feedback
- •Side projects as a strategy to expand creativity and connect dots
- •Use prototypes to uncover what’s possible and refine product direction
- 33:04 – 34:15
Framework 1: Anatomy of an agent—models, tools, and memory/personalization
Jacqueline outlines a practical mental model for agents: pick the right model capabilities, combine them with tool use, and design memory intentionally. She references browsing agents (UI actions), API/MCP integration, and the product-driven question of what the agent should remember to achieve user success.
- •Model selection by needed modalities (text/image/audio/video/code)
- •Tool use as a power multiplier: search APIs, UI actions, MCP, integrations
- •Memory/personalization as a product decision tied to goals and success metrics
- •Example context: browsing agent work (Project Mariner) emphasizes UI actions
- 34:15 – 36:35
Framework 2: User Interaction Spectrum—‘do it for me’ vs ‘do it with me’
This chapter focuses on UX design choices for AI products based on how autonomous the experience should be. Jacqueline contrasts hands-off agents (deep research, audio overview generation) with collaborative modes (vibe coding, interactive audio overviews) and explains how the spectrum affects product design.
- •Two ends: autonomous task execution vs collaborative co-pilot
- •Examples of ‘do it for me’: deep research; generating audio overviews from sources
- •Examples of ‘do it with me’: vibe coding; interactive interruptions/Q&A in audio overviews
- •Design implication: user involvement level shapes flow, prompts, and UI
- 36:35 – 39:51
Framework 3: The Inverted Triangle—think big, ship fast via MVP scope, positioning, and audience
Jacqueline argues that AI products must start with ambitious visions to avoid rapid commoditization, then narrow to ship quickly. She shares three levers for early shipping: cut MVP scope, position as beta/experiment, and control rollout audience with trusted testers or waitlists.
- •Think big to avoid building soon-to-be-commoditized features
- •Ship sooner by cutting scope to true MVP essentials
- •Use positioning labels (beta/experiment/concept) to set expectations
- •Control audience: internal dogfooding → trusted testers → broader release
- •Examples: Opal and Mixboard launched before full integration breadth
- 39:51 – 51:50
Paradigm shift + future-proofing questions (and first-principles thinking)
Jacqueline offers two key product questions: are you creating a new workflow (car) or just optimizing an old one (faster horse), and what happens when models get better. She connects this to first-principles thinking, second-order thinking (platforms over one-off apps), and being willing to discard work when the tech frontier changes.
- •Paradigm shift test: new workflow vs incremental process improvement
- •Future-proofing: will model updates commoditize your core feature or unlock more?
- •Example: Mixboard rethought image editing once Nano Banana changed the game
- •First-principles framing: focus on true user need (e.g., “bring memories to life”)
- •Second-order thinking: build platforms/tools (e.g., Opal) vs single niche apps
- •Magic-wand planning: identify human steps that exist only due to model limits
- 51:50 – 57:47
Hiring for AI PM: the 6 characteristics Google looks for
Jacqueline shares what she prioritizes when hiring AI PMs: taste, visionary systems thinking, operating in ambiguity, storytelling, execution ownership, and AI intuition/creativity. She emphasizes idea generation as an ongoing skill because commoditization can happen quickly in AI.
- •Exceptional product taste and user-centric craft
- •Visionary leadership + systems thinking to predict what’s next
- •Clarity in chaos + empathetic resolve in ambiguous zero-to-one work
- •Compelling storytelling when data is limited
- •Full-spectrum execution as role boundaries blur
- •Deep AI intuition + applied creativity; generate many strong ideas continuously
- 57:47
AI PM resume + Google interview process + an 18-month roadmap to break in
Jacqueline explains how to build an AI PM resume that’s concise, specific, and proof-driven, including links to side projects and public work. She addresses interview expectations (avoid jumping straight into vibe coding; clarify approach), outlines a common Google loop variant, and closes with a practical 18-month plan: build, share publicly, network, and immerse in the ecosystem.
- •Resume: 1 page, show-don’t-tell, link examples, careful design, proofread, add context for impact
- •Highlight above-and-beyond work (side projects, hackathons, talks) to compensate for limited AI-native job experience
- •Interview advice: approach as product/design; propose your plan and confirm expectations
- •Process notes: recruiter screen → written questions (for her roles) → 45-min call → 4-round loop; sometimes no end-stage team match
- •18-month roadmap: build/create artifacts, build in public, network/events, learn via courses/podcasts/Substacks, adapt PM fundamentals
Why Google is suddenly leading in AI—and what you’ll learn in this “AI PM masterclass”
Aakash frames Google’s momentum in AI models and tooling (image, video, and workflow builders) and introduces the goal: an insider-level breakdown of how to build AI products and become an AI PM. Jacqueline Kunzelmann joins to share demos, frameworks, and career advice for breaking into AI PM roles.
- •Google’s recent AI momentum and model/tool lineup (image, video, workflows)
- •What an “AI PM masterclass” will cover: tools, product building, hiring, and roadmap
- •Setting expectations: practical demos + strategic frameworks + career guidance
Demo: Nano Banana image editing—capabilities tour and ‘world model’ intuition
Jacqueline shares a rapid tour of Nano Banana’s image manipulation strengths, from object rotation and annotations to sketch-to-art transformations. She highlights the model’s ability to infer realistic changes (e.g., seasons by location), signaling a deeper world understanding that expands product possibilities.
- •Core editing tricks: rotate objects, add infoboxes, transform sketches
- •Style transfer and rapid creative generation for non-designers
- •World-model inference example: “winter” looks different in Toronto vs. San Francisco
- •Multimodal convergence expands what “image editing” can mean
Where to access Nano Banana: AI Studio vs Gemini app vs Mixboard canvas
Jacqueline explains the practical access paths for Nano Banana and why output quality can vary by surface. She introduces Mixboard as a canvas-style UI that supports brainstorming, batch transformations, and visual ideation beyond the chat interface.
- •Two primary access points: AI Studio and Gemini app
- •Mixboard: open-ended canvas for creative/visual workflows
- •Batch/group uploads and bulk transformations (e.g., convert multiple images to sketches)
- •Choosing surface by workflow: iterate and compare results across tools
