Aakash Gupta

Gemini Gems Masterclass with the Creator at Google: 3 Gems You Must Build

Lisa Huang (SVP Product at Xero, creator of Gemini Gems at Google) breaks down how to build personalized AI with Gemini Gems, what it takes to ship accurate agents in fintech, and how to navigate your AI PM career - from Apple to Meta to Google to Xero. Full Writeup: https://www.news.aakashg.com/p/lisa-huang-podcast Transcript: https://www.aakashg.com/gemini-gems-meta-ray-ban-ai-and-building-agents-at-scale-with-lisa-huang/ --- Timestamps: 0:00 - Intro 1:49 - Guest introduction 2:05 - What are Gemini Gems? 3:52 - The 3 must-have Gems for every PM 6:05 - Live demo: building a product strategy Gem 9:01 - The story behind building Gemini Gems 10:32 - Ads 11:39 - Gemini Gems vs ChatGPT custom GPTs 16:45 - Career lessons from Apple, Meta and Google 23:05 - Building the AI assistant for Meta RayBan glasses 27:43 - Introducing JAX - Xero's financial super agent 32:22 - How to measure an AI agent 37:39 - Will AI replace PMs? 38:40 - Ads 39:45 - Breaking into AI PM 51:15 - Outro --- 🏆 Thanks to our sponsor - Reforge Build: AI prototyping built for product teams - https://reforge.com/aakash --- Key Takeaways: 1. Stop briefing your LLM from scratch every time - Gemini Gems hold your context permanently. Your role, your company strategy, your writing style. Build it once and it already knows everything the next time you open it. 2. Every PM needs 3 Gems - A writing clone trained on your PRDs and emails. A product strategy advisor loaded with your company docs and competitor analysis. A user research synthesizer that ingests raw transcripts and surfaces key themes. 3. Vague instructions are the number one mistake - "Help me write better" gets you nothing. Write a full page of context. Your role, your audience, your format preferences. The more specific, the more personalized the output. 4. Gemini Gems vs ChatGPT custom GPTs - OpenAI framed GPTs as an app store ecosystem. Google focused on personal productivity instead. First principles beat copying a competitor's framing, and the GPT store never took off. 5. On-device AI is the future for wearables - Cloud is the default today but once a device is on your face all day, people want their data staying local. Privacy beats performance when the device is that personal. 6. Accuracy is the product in high-stakes AI - LLMs out of the box are not great at math, accounting, or tax. Winning agents combine deep domain knowledge with proprietary data that no general-purpose model can access. 7. Measure agents in three layers - Quality first (evals, human annotators, LLM judges). Product metrics second (adoption, retention, CSAT). Business impact third (revenue attribution, ARR). Skip to layer three without the foundation and you are measuring on sand. 8. AI will not replace PMs - it will replace the execution work. Writing PRDs, creating mocks, managing roadmaps. What stays is product judgment. The ability to look at ambiguous signals and say this is the right bet and here is why. 9. The PM role is becoming a hybrid - PM to engineer ratios will compress. The expectation is that PMs also build. Not just spec and hand off, but prototype, design, and code enough to show what they mean. The tools to do this exist right now. 10. Your company's permission is not required - Most companies are not fine-tuning models. They are using the same consumer tools you already have. Build Gems. Build projects. Build small AI products with your personal data. There is nothing stopping you. --- 👨‍💻 Where to find Lisa Huang: LinkedIn: https://www.linkedin.com/in/lisaxhuang/ Xero: https://www.xero.com/us/ai-in-accounting/jax/ 👨‍💻 Where to find Aakash: Twitter: https://www.x.com/aakashg0 LinkedIn: https://www.linkedin.com/in/aakashgupta/ Newsletter: https://www.news.aakashg.com #gemini #aipm --- 🧠 About Product Growth: The world's largest podcast focused solely on product + growth, with over 200K+ listeners. 🔔 Subscribe and turn on notifications to get more videos like this.

AGAakash GuptahostLHLisa Huangguest

Mar 5, 2026·52m·Watch on YouTube ↗

📖 CHAPTERS

1. 1

   Why Gemini Gems matter: personalized context for better outputs

   Aakash frames Gemini’s rise and tees up why using Gemini Gems unlocks meaningfully better results than prompting a generic LLM each time. Lisa explains the core problem: LLMs are powerful but lack persistent context, forcing repeated re-explanations.

2. 2

   The 3 must-have Gems for product managers

   Lisa outlines three core Gems every PM should create to cover the most common PM workstreams. Each Gem corresponds to a major PM responsibility: communication, strategic decision-making, and research synthesis.

3. 3

   How to build a Gem: instructions, knowledge, and iteration loop

   Lisa breaks down the simple creation flow inside Gemini and what actually drives quality. The “build” process is framed like creating a mini-product: specify behavior clearly, ground it in the right documents, then test and refine.

4. 4

   Live demo: building a Product Strategy Buddy Gem

   Lisa screenshares and constructs a product strategy Gem in real time, showing how it’s configured and validated. She demonstrates adding company context (strategy, roadmap, competitive teardown) and running a first test prompt to evaluate output quality.

5. 5

   Team sharing and organizational use cases

   The conversation shifts from individual productivity to team leverage. Lisa explains that Gems can be shared across colleagues, especially when a team has common context and repeatable workflows.

6. 6

   Origin story: why Google built Gemini Gems

   Lisa recounts the 2023 product insight behind Gems: users struggled to discover and repeatedly access Gemini’s capabilities. Personas and customizable assistants were a way to improve discoverability, fit mental models, and enable sharing.

7. 7

   Gemini Gems vs ChatGPT custom GPTs: different product thesis

   Aakash asks how Gems compare to OpenAI’s custom GPTs and whether Gemini copied the concept. Lisa explains timing overlap and highlights a strategic divergence: Gemini focused on productivity rather than an “app store” monetization ecosystem.

8. 8

   PM Gem portfolio thinking: mapping Gems to core PM skill areas

   They discuss how to decide what to build beyond the three starter Gems. Lisa proposes breaking PM work into strategy, execution, communication, and research signals—then creating multiple specialized Gems across those buckets.

9. 9

   Career philosophy and lessons from Apple, Meta, and Google

   Lisa transitions into career advice, emphasizing curiosity as the guiding principle rather than a rigid plan. She compares product cultures across Apple, Meta, and Google and extracts what she carried forward from each environment.

10. 10

    Building the Meta Ray-Ban glasses AI assistant: constraints and zero-to-one reality

    Lisa details her work on the first-gen assistant for Ray-Ban Stories (2019–2021), including skepticism she faced and how the project got funded. She highlights the unique challenges of wearable AI: privacy, hardware constraints, and partnership complexity.

11. 11

    Cloud vs on-device AI for wearables and how to build in fast-changing tech

    They explore architectural tradeoffs for AR AI and what’s likely to change. Lisa predicts a wave toward on-device AI driven by privacy and practicality, and advises PMs to balance deep tech understanding with user value and rapid iteration.

12. 12

    Xero’s JAX financial super-agent: automating workflows with domain + data advantage

    Lisa introduces Xero and explains JAX as an umbrella initiative to map financial workflows and automate jobs-to-be-done with agents. She argues Xero’s differentiation comes from deep workflow knowledge and rich transaction-level data across SMBs.

13. 13

    Reliability in high-stakes agents: hybrid systems, expert annotation, and quality flywheels

    Aakash probes hard lessons in building agents where accuracy is critical (finance). Lisa explains why raw LLMs aren’t enough and how Xero uses a hybrid approach with programmatic controls, evaluations, and domain expert oversight.

14. 14

    Measuring an AI agent: quality → engagement → business impact

    Lisa provides a measurement framework that ladders from technical correctness to adoption and finally monetization. She describes how eval criteria vary by use case and why quality tracking must be ongoing as the product evolves.

15. 15

    Will AI replace PMs? Role compression, PM-as-builder, and how to break into AI PM

    Lisa argues PMs won’t be replaced because judgment and taste remain essential, but team structures and ratios will compress. She advises PMs to become builders—using AI to prototype, design, and even code—and shares a practical roadmap for breaking into AI PM roles.