Aakash GuptaGoogle AI PM Reveals the Tools 99% of Product Managers Don’t Use
CHAPTERS
AI prototyping as a PM superpower + what “AIPM tool hopping” looks like
Aakash introduces Marily Nika (Google AI PM) and frames the episode around the underrated AI tools that dramatically speed up PM work. Marily outlines her everyday stack and the concept of “tool hopping” to move from idea → artifact → prototype → research faster than traditional workflows.
- •Why AI tools now meaningfully change PM speed and quality of output
- •Marily’s core daily stack: AI Studio, Opal, NotebookLM, Perplexity, ChatGPT/Gemini, Fireflies
- •Different “hats” (day job vs teaching/bootcamp) drive different tool choices
- •Theme: prototypes and artifacts become starting points, not final deliverables
Why these 6 tools matter: mapping tools to the PM lifecycle
Marily explains the specific PM jobs each tool helps with—prototyping, mini-app workflows, domain learning, user voice research, writing artifacts, and meeting capture. The focus is on picking a small set of high-leverage tools rather than an overwhelming tool zoo.
- •AI Studio: rapid iteration and visualization of product ideas
- •Opal: quick mini-app creation and workflow-style automation
- •NotebookLM: grounded learning/research on only the sources you provide
- •Perplexity: fast pulse-check of real user opinions (especially Reddit)
- •ChatGPT/Gemini: custom generators for PRDs/PRFAQs and other artifacts
- •Fireflies: cross-platform meeting notes and follow-ups
Google AI Studio walkthrough: turning a plain-English idea into an app prototype
Marily demos AI Studio’s prototyping interface and highlights features PMs should use: model choice (Pro vs Flash), file upload, speech-to-text, and the inspiration gallery. The point is to get from concept to something visual and discussable in minutes.
- •AI Studio UI basics: prompt area, model selection, attachments, speech-to-text
- •Using the gallery to get inspiration and patterns from other builders
- •Example prototype inspiration (e.g., Windows 95-style app)
- •Choosing the right model for speed vs capability
Hands-on prototype: NanoBanana-powered collage generator for LinkedIn content
They build a prototype concept: upload a user photo + bucket-list goals → generate a collage suitable for LinkedIn. The demo emphasizes how quickly PMs can validate an interaction flow and output style without engineering help.
- •Defining inputs/outputs clearly: photo + bucket list items → collage
- •Why the image model choice matters (NanoBanana for text-to-image)
- •Iterating via natural-language tweaks to UI and behavior
- •Using prototypes to clarify product vision early
AI prototyping deep dive: when prototypes replace early docs (and when PRDs still matter)
Marily explains how her workflow has shifted from ‘idea → PRD → alignment’ to ‘idea → prototype → bring engineering/science in.’ She also notes that documentation still matters for complex, cross-functional, async-heavy initiatives.
- •New workflow: prototype first, then align stakeholders on the artifact
- •Prototypes make vision concrete and reduce persuasion/interpretation costs
- •PRDs remain important for complex, global, multi-team execution
- •Startup vs big-tech differences in how “necessary” PRDs are
Design-system matching in AI Studio: making prototypes look like your product
Aakash raises a common PM problem: getting AI prototypes to follow an existing design system. Marily shares a simple approach—upload a screenshot and instruct the tool to mirror the visual language—highlighting how much better this has gotten recently.
- •Use screenshots as visual constraints for colors/typography/layout
- •Prompting specific design tweaks (contrast, font, tone) works well
- •AI Studio responsiveness to visual iteration vs other tools
- •Why “presentation quality” impacts stakeholder alignment
Opal mini-apps: building an automated workflow from one sentence
Marily introduces Opal (Google Labs) and recreates the same collage use case. Opal generates a multi-step workflow (Zap-like) and an app experience, turning a short instruction into an expanded prompt and runnable mini-app.
- •Opal’s two modes: natural language and drag-and-drop workflow blocks
- •Automatic expansion of a short prompt into a detailed system prompt
- •Workflow structure: inputs → image generation → HTML collage output
- •Tradeoff: fast experimentation vs less direct code portability
Opal vs AI Studio: strengths, limitations, and best-fit PM use cases
They compare the tools: AI Studio for more direct prototyping feel and iteration; Opal for quick workflows and templated mini-apps. Marily recommends caution making Opal a core workflow since it’s experimental and can be inconsistent.
- •Opal can be slower/quirkier and may need more iterations
- •AI Studio feels better for precise UI/visual tweaks and faster iteration
- •Opal shines for lightweight “generate X” workflows (e.g., empathy maps)
- •Guidance: don’t rely on experimental tools for mission-critical workflows
NotebookLM for domain expertise: grounded research + reusable ‘personalized assistant’
Marily explains NotebookLM as a research assistant grounded only in provided sources (PDFs, videos, Drive, notes). She uses it to judge bootcamp demo-day pitches by ingesting audio clips and generating an audio overview that selects winners by criteria.
- •Key differentiator: it’s grounded in your uploaded sources (not the open web)
- •Supports many formats: PDFs, YouTube, Drive sync, audio/video sources
- •Demo-day use case: evaluate pitch creativity/impact/storytelling
- •Audio Overview + interactive Q&A makes outputs engaging and memorable
NotebookLM in action: interview prep, summarization at scale, and UXR synthesis
Marily shares additional high-impact PM uses: learning a new domain quickly, summarizing long investor relations videos for interviews, and distilling insights from massive UXR libraries. The emphasis is on extracting what matters and ignoring irrelevant content.
- •Interview prep: ingest JD + long video → extract key points for the role
- •New domain ramp-up: load books/videos and ask targeted learning questions
- •User research at scale: synthesize patterns across hundreds of interviews
- •‘Read between the lines’ to interpret user sentiment and priorities
Perplexity for Reddit research: capturing the ‘voice of the people’ for MVP decisions
Marily shows how she uses Perplexity’s discussions/opinions filter to search Reddit rather than the whole web. She uses it to validate demand and turn real user opinions into a prioritized MVP feature list.
- •Use Perplexity filters to focus on discussions/opinions (Reddit-heavy)
- •Start with a market question, then drill into sourced threads
- •Prompt follow-up: generate MVP feature recommendations from findings
- •Outcome: faster product sense-making and early PMF signal detection
ChatGPT (and custom GPTs) as an artifact engine: PRD generator + PRFAQ practice
Marily demonstrates a custom PRD generator trained on her voice and preferred structure. They discuss when AI-written artifacts help vs create ‘slop,’ and she explains PRFAQ as a press-release-first method to clarify the end state and work backward.
- •Custom GPT asks clarifying questions before generating the PRD
- •PRD output sections: personas, use cases, AI features, tech stack, research
- •Workflow: paste outputs into Docs or feed into prototyping tools
- •PRFAQ vs PRD: vision/press-release-first vs requirements-focused spec
Common AI mistakes + meeting capture: normalize AI use and automate recall
Marily’s biggest caution is cultural, not technical: don’t hide that you used AI. She then explains her note-taking setup (Fireflies + Gemini) and why cross-platform meeting coverage matters for busy PMs.
- •Biggest mistake: being embarrassed about using AI—be transparent
- •AI advantage: speed + iteration; others using tools will outpace you
- •Fireflies joins Zoom/Meet broadly; Gemini is preferred on Google Meet
- •Meeting notes as a force multiplier for follow-through and accountability
18-month AIPM roadmap + interview red flags + the future of PM
Marily outlines how to move into AI PM via adjacent moves (‘be a crab’), becoming AI-literate, and understanding coding fundamentals without necessarily coding. She shares interview red flags (solution-first, weak PM craft, confusing PM vs program) and argues PM judgment and strategy remain essential even as AI becomes ubiquitous.
- •Career strategy: move adjacent; leverage domain expertise as a differentiator
- •AIPM readiness: AI literacy, probabilistic mindset, data dependency awareness
- •Coding guidance: understand APIs, version control, productionization challenges
- •Interview red flags: skipping the why/who/success metrics; PM vs TPM confusion
- •Future view: AI becomes embedded in most products; PM craft (judgment/strategy) stays vital
