Aakash GuptaGemini Gems Masterclass with the Creator at Google: 3 Gems You Must Build
Aakash Gupta and Lisa Huang on build Gemini Gems, ship AI agents, and future-proof PM careers.
In this episode of Aakash Gupta, featuring Aakash Gupta and Lisa Huang, Gemini Gems Masterclass with the Creator at Google: 3 Gems You Must Build explores build Gemini Gems, ship AI agents, and future-proof PM careers Gemini Gems are customizable Gemini instances that retain instructions and uploaded context so users stop re-prompting and can get consistently tailored outputs.
At a glance
WHAT IT’S REALLY ABOUT
Build Gemini Gems, ship AI agents, and future-proof PM careers
- Gemini Gems are customizable Gemini instances that retain instructions and uploaded context so users stop re-prompting and can get consistently tailored outputs.
- Lisa Huang recommends three must-have PM Gems—a writing clone, product strategy advisor, and user research synthesizer—mapped to core PM responsibilities of communication, strategy, and insight generation.
- Effective Gems require detailed instructions, the right contextual “knowledge” files, specialization by job-to-be-done, and ongoing iteration like a mini product.
- In high-stakes domains like finance, successful AI agents require hybrid systems (LLMs plus programmatic controls), strong domain workflows, robust evals, and human/LLM-judge quality loops.
- PM careers won’t be replaced by AI, but team structures and expectations will change toward “PMs as builders,” making AI tool fluency, prototyping, and technical depth increasingly mandatory.
IDEAS WORTH REMEMBERING
9 ideasTreat a Gem like a reusable context container, not a one-off prompt.
Gems exist because LLMs lack persistent context; by storing instructions and curated documents, you reduce repetitive prompting and get more reliable, “you-shaped” outputs over time.
Every PM should start with three Gems tied to core PM work.
A writing clone speeds stakeholder communication, a product strategy advisor supports decision-making with company/market context, and a user research synthesizer turns raw feedback into actionable insights.
High-performing Gems are built with specificity, examples, and scoped jobs.
Vague prompts (“help me write better”) underperform; detailed instructions plus relevant artifacts (PRDs, emails, research transcripts) and specialized Gems per task produce better consistency.
Iteration is not optional—quality comes from a feedback loop.
Lisa frames Gems as mini products: test outputs, refine instructions, update knowledge files as reality changes, and keep tightening until the Gem reliably matches your intent.
Gems differ from custom GPT positioning: prioritize productivity over monetization.
Google saw instructions as easily copyable and focused less on an “app store” model, instead optimizing for personal/team amplification and sharing within shared-context environments.
In finance, “accuracy to the decimal” forces hybrid agent designs.
LLMs alone are weak at math/accounting, so Xero’s JAX layers domain workflow understanding, proprietary transaction data, programmatic checks, and robust eval systems to ensure reliability.
Measure AI agents in three stacked layers: quality → adoption → business impact.
Start with baseline correctness via evals (human + LLM judges), then track product metrics (usage/retention/CSAT), and finally attribute monetization or revenue impact once the first two are strong.
AI won’t replace PM judgment, but it will compress teams and raise the bar.
Lisa argues PM value is judgment under ambiguity, yet PM/eng ratios will tighten and PMs will be expected to prototype/design/build (often with AI) rather than only write specs.
To break into AI PM, don’t wait for permission—build evidence of obsession.
Use consumer AI tools on personal projects, show deep user/domain research initiative (like the TikTok small-business synthesis example), and practice standardized big-tech interviews repetitively.
WORDS WORTH SAVING
5 quotesGemini Gems are custom versions of Gemini that you can create for your specific use case.
— Lisa Huang
If you aren't using Gemini Gems, you aren't getting the most out of Gemini.
— Lisa Huang
Let’s call them the writing clone, the product strategy advisor, and the user research synthesizer.
— Lisa Huang
Accuracy is not a nice-to-have, it’s a core part of our differentiation.
— Lisa Huang
I pay them for their product judgment.
— Lisa Huang
QUESTIONS ANSWERED IN THIS EPISODE
5 questionsFor each of the three “must-have” PM Gems, what are the minimum viable instruction blocks and the best 3–5 example artifacts to upload to get strong results fast?
Gemini Gems are customizable Gemini instances that retain instructions and uploaded context so users stop re-prompting and can get consistently tailored outputs.
When you say Gems “lack persistent memory” beyond instructions and files, what’s the practical maintenance routine (weekly/monthly) to keep a Gem current without it drifting?
Lisa Huang recommends three must-have PM Gems—a writing clone, product strategy advisor, and user research synthesizer—mapped to core PM responsibilities of communication, strategy, and insight generation.
If instructions are easily copied, what product moats (if any) could a Gem ecosystem have built—distribution, data, workflow integrations, or something else?
Effective Gems require detailed instructions, the right contextual “knowledge” files, specialization by job-to-be-done, and ongoing iteration like a mini product.
In the Ray-Ban assistant work, what were the highest-impact product decisions that were constrained by battery, latency, and privacy, and how did you trade them off?
In high-stakes domains like finance, successful AI agents require hybrid systems (LLMs plus programmatic controls), strong domain workflows, robust evals, and human/LLM-judge quality loops.
For finance agents like JAX, what specific failure modes show up most often (math errors, tool-calling, policy, user intent), and what mitigations worked best?
PM careers won’t be replaced by AI, but team structures and expectations will change toward “PMs as builders,” making AI tool fluency, prototyping, and technical depth increasingly mandatory.
EVERY SPOKEN WORD
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome