Aakash GuptaGoogle AI PM Reveals the Tools 99% of Product Managers Don’t Use
CHAPTERS
- 0:00 – 6:41
Six-tool AIPM stack overview: prototyping, mini-apps, research, artifacts, and meeting notes
Marily Nika and Aakash Gupta set up the episode’s premise: most PMs aren’t using a small set of high-leverage AI tools that can dramatically speed up discovery, communication, and execution. Marily names her daily drivers—AI Studio, Opal, NotebookLM, Perplexity (Reddit mode), custom GPTs/Gems for PRDs, and Fireflies—and frames them as “tool hopping” across different PM hats.
- •AI Studio for rapid AI prototyping and iteration
- •Opal (Google Labs) for natural-language mini apps and workflow-like automations
- •NotebookLM for domain learning and context-bounded analysis
- •Perplexity with social/Reddit filtering for user voice and market signals
- •Custom GPTs/Gems for PRDs/PRFAQs and Fireflies for meeting notes/follow-ups
- 6:41 – 8:44
Google AI Studio walkthrough: interface, inspiration gallery, and NanoBanana model selection
Marily demonstrates the AI Studio prototyping environment, highlighting model selection (2.5 Pro vs Flash), file upload, speech-to-text, and a gallery for inspiration. She showcases an example app (“Dreamy 1995”) and explains why NanoBanana is her go-to text-to-image model when prototypes need strong image generation.
- •AI Studio basics: idea prompt area, model picker, uploads, speech-to-text
- •Gallery browsing to spark ideas and reduce blank-page friction
- •Example: Windows 95-style nostalgic prototype built via natural language
- •NanoBanana positioned as a standout text-to-image model
- •Early emphasis on fast iteration and visual communication
- 8:44 – 16:33
Live prototype build: LinkedIn-ready bucket-list collage generator (and what breaks in real demos)
They build a prototype concept: upload a user photo, collect bucket-list scenarios, and generate a collage suitable for LinkedIn content. The demo illustrates both the speed of first iterations and real-world constraints like API quota limits during live builds, prompting a pivot to discussing process rather than perfect output.
- •Prototype concept: photo input + bucket-list prompts → generated collage
- •Prompting to align the tool to an image-heavy use case (NanoBanana)
- •First-iteration usability: upload, edit, then generate
- •Observed differences in default aesthetics vs other tools (Gemini vs Claude vibe)
- •Quota/API limits as practical constraints; knowing when to move on
- 16:33 – 18:41
When to prototype vs write a PRD: shifting the product workflow toward visual alignment
Marily explains how prototyping increasingly precedes documentation: instead of ‘idea → PRD → comments,’ she prototypes first and then brings engineers/scientists in to debate a concrete artifact. She clarifies when PRDs still matter—complex, cross-functional, async-heavy efforts—while noting startups can often move faster with prototypes.
- •New workflow: idea → prototype → pull in XFN partners for feedback
- •Prototypes improve persuasion and shared understanding of end behavior
- •PRDs still valuable for complex, global, async collaboration
- •Startup vs big-tech documentation needs are diverging
- •Prototypes make PM work more creative and less purely document-driven
- 18:41 – 19:40
Design-system matching in AI Studio: screenshots as style anchors + rapid visual tweaks
They address a common PM question: how to get AI prototypes to match an existing design system. Marily recommends providing screenshots as visual references and iterating via simple design instructions (contrast, fonts, background), noting that modern tools converge on desired styles faster than earlier generations.
- •Upload screenshots to anchor colors/layout to an existing product
- •Direct natural-language tweaks for contrast, backgrounds, fonts
- •AI Studio responsiveness to visual adjustments as a differentiator
- •Personalization for specific audiences (e.g., kid-friendly apps)
- •Practical: add example prompts and image-generation examples to improve outcomes
- 19:40 – 22:59
Opal mini-apps: natural-language to drag-and-drop workflows (Zap-like)
Marily introduces Opal as a Google Labs experiment for quickly generating mini apps and step-based workflows from plain English. She recreates the collage-generator idea in Opal, showing how a single sentence expands into a structured prompt + generation steps and produces an app-like interface rather than editable deployable code.
- •Opal positioning: ‘AI mini apps’ with workflow nodes and a runnable UI
- •Natural-language spec expands into a detailed internal prompt
- •Outputs an Opal-hosted app experience, not portable code
- •Themes/styles can be selected to influence output
- •Tradeoff: speed of app creation vs flexibility of code ownership
- 22:59 – 27:03
Opal iteration: debugging wonky outputs and where workflows shine for PM work
The first Opal run produces repetitive images, prompting Marily to show how iteration and better input structure fixes results. She highlights additional PM-friendly workflows like instantly generating empathy maps and other lightweight artifacts—while cautioning that Opal is experimental and shouldn’t be a core dependency until tested.
- •Common failure mode: repeated/undifferentiated outputs across images
- •Fixes via restructuring inputs (separate prompts per image) and iterating
- •Demonstration of a refined version that produces better collage diversity
- •Good ‘quick win’ use cases: empathy maps and lightweight automations
- •Caution: experimental product—avoid over-reliance without validation
- 27:03 – 30:59
NotebookLM as a context-bounded research assistant: bootcamp ‘AI judge’ demo
Marily explains NotebookLM’s core advantage: it only uses the sources you provide, enabling controlled, domain-specific reasoning. She shows a creative use case—uploading audio clips of student pitches and generating an ‘Audio Overview’ that acts like a panel of judges selecting winners based on criteria like creativity and storytelling.
- •NotebookLM ingests PDFs, notes, YouTube, Drive folders, and audio
- •Key differentiator: grounded on user-provided sources (less free-roaming)
- •Audio Overview creates an engaging, human-like ‘host’ narration
- •Bootcamp demo: AI judges evaluate pitches and pick winners
- •Extra learning tools: mind maps, flashcards, interactive Q&A modes
- 30:59 – 37:28
NotebookLM for rapid domain mastery and interview prep (and scaling UXR synthesis)
Beyond the demo, Marily shares high-impact PM applications: mastering unfamiliar domains quickly and preparing for interviews by extracting the most relevant points from long videos and documents. She also positions NotebookLM as a superpower for synthesizing massive qualitative research—hundreds of pages and hours of interviews—into actionable product insights.
- •Interview prep: combine job description + investor relations video → key takeaways
- •Fast onboarding into new domains (e.g., healthcare) via curated source bundles
- •Guiding prompts to ignore irrelevant content and focus on outcomes
- •UXR synthesis: distill themes and ‘read between the lines’ across many interviews
- •NotebookLM as a scalable context engine for PM judgment and prioritization
- 37:28 – 39:50
Perplexity for ‘voice of Reddit’: social-only research → MVP feature list
Marily demonstrates Perplexity’s ‘discussions and opinions’ filter to search Reddit instead of the broader web. She uses it to validate interest in a product idea and then pushes it further—having Perplexity derive must-have MVP features based on aggregated Reddit sentiment, with direct links to sources for verification.
- •Turn off web search; focus on discussions/opinions for authentic user voice
- •Example query: interest in a step-tracking ring for young professionals
- •Rapid aggregation of many Reddit threads with citations and links
- •Follow-up prompt to convert findings into MVP feature requirements
- •A focused, differentiated Perplexity use case vs general Q&A
- 39:50 – 43:38
Custom ChatGPT PRD generator: voice-trained artifacts + PRFAQ vs PRD
Marily shares her heavily trained custom GPT that generates PRDs in her voice for both 0→1 and 1→1 work, and explains how she uses outputs as inputs to docs or prototyping tools. She also explains PRFAQ—a press-release-first practice inspired by Amazon—that helps teams visualize the end state and work backwards.
- •Custom GPT probes with PM-style questions before drafting the PRD
- •End-to-end PRD sections: personas, features, AI use cases, tech stack, research
- •Workflow: paste in research outputs (e.g., Perplexity features) → generate PRD
- •PRFAQ vs PRD: end-state press release + FAQs vs detailed requirements
- •Practical gap: ‘last mile’ handoff into Docs/tools still clunky
- 43:38 – 48:55
Using AI for artifacts without ‘slop’: editing strategy and normalization at work
They debate whether AI-generated PRDs dilute thinking; Marily argues even a mediocre draft beats a blank page and saves time when tuned properly. Her strongest cultural advice: don’t hide AI usage—share it openly—because the real risk is being outpaced by PMs who adopt these tools while others resist.
- •‘Writing is thinking’ concern addressed via better prompt training and curation
- •AI drafts as scaffolding: faster titles, structure, and baseline completeness
- •Normalize AI usage; don’t be embarrassed or conceal it
- •Custom GPTs/Gems for repeatable artifacts (PRD, PRFAQ) vs inconsistent reports
- •Competitive threat framing: people using tools replace those who don’t
- 48:55 – 51:15
Fireflies note-taking and the ‘human-in-the-loop’ MVP story + choosing note-takers
Marily explains she uses Fireflies primarily outside Google because it works across Zoom and Google Meet, ensuring coverage across varied calendars and meeting platforms. They discuss the infamous Fireflies story where humans initially took notes behind the scenes—an example of scrappy MVP validation—and compare Fireflies with Gemini for Google Meet notes.
- •Cross-platform reliability: Fireflies joins Zoom, Meet, and more automatically
- •Use-case split: Gemini for Google Meet; Fireflies for everything else
- •Notetakers as productivity + follow-up assurance tools
- •‘Human pretending to be AI’ as a bootstrapped MVP tactic (and its risks)
- •Practical selection criteria: coverage across meeting tools and workflows
- 51:15 – 55:16
Experimentation cadence + future of PM: judgment, strategy, and getting tool access
Marily recommends blocking one hour weekly for experimentation as tools and workflows change rapidly (e.g., new AI browsers like OpenAI’s Atlas and Perplexity’s Comet). She argues PM craft remains essential—defining success, strategy, and judgment—and advises PMs without tool access to pitch leadership with a one-page ROI case for modernization.
- •Calendar habit: 1 hour/week dedicated to trying new tools
- •Stay informed via newsletters and creators; expect constant workflow shifts
- •PMs won’t be replaced by AI—judgment, strategy pivots, and morale context matter
- •How to get access: write a one-pager with use case + hours saved + ROI
- •Reframe lack of tools as an internal leadership opportunity
- 55:16 – 1:02:10
18-month AIPM roadmap: ‘be a crab,’ build AI literacy, and interview red flags
Marily outlines how aspiring AIPMs can transition by moving adjacent to their existing expertise (‘be a crab’) and leveraging domain advantage to stand out. She covers whether AIPMs must code (understanding fundamentals matters more) and flags common interview mistakes: over-indexing on AI knowledge while neglecting PM craft, and confusing product management with program management.
- •Career strategy: lateral moves that capitalize on domain adjacency (hearing aids → AirPods)
- •AI builder vs AI experiences PM roles: platforms/APIs vs end-user product experiences
- •Coding guidance: understand APIs, version control, productionization challenges
- •Interview red flags: solution-first without user/why/success metrics; ‘AI-first’ but not product-first
- •Another red flag: conflating PM with TPM/program execution focus
- 1:02:10 – 1:06:57
Is ‘AIPM’ a useful label? Why AI will permeate most products + wrap-up
They discuss criticism of the AIPM title and Marily’s view that it’s useful today but will blur into ‘PM’ as AI becomes embedded across products. She argues nearly any product can incorporate smart functionality and closes with encouragement to trial the demonstrated tools to stay competitive.
- •AIPM label: helpful signal now; likely merges into general PM over time
- •Subcategories exist but may blur as AI becomes ubiquitous
- •Argument: most products will include ‘smart’ features or AI-enabled insights
- •Practical next step: try at least one tool from the episode and build a habit
- •Episode close: recap and calls to subscribe/follow
