Inside Gemini: How Google Runs Product for Its Model

In this episode, Tulsee Doshi, who leads product for Google's Gemini models, joins me for the fourth conversation in the "Inside PM" series. Tulsee has spent eleven years at Google, moving through Search and YouTube before taking on product for the Gemini family and Google's video, image, and audio models. We get into how her team avoids building a boring product, why their blend of platform work and lab work feels like a new kind of product management, and what she looks for in the PMs she thinks will thrive in that environment.–Key topics:• Why customer requests get weighed against research the customers can't see• Why stack-ranking every request would leave you with a boring product• The 20 prompts trick for turning a vague ask into a great eval• How to know a model is ready to ship without waiting a year• Reading a model's "vibes," and why Simon Tokumine's gut check carries real weight• Prototyping features that don't work yet on purpose• Why "technical" on her team means thorough, not a computer science degree• The Nano Banana moment that "fuels all of the other chaos"–Timestamps:00:00 Introduction03:28 Welcoming Tulsee Doshi to break down how Google builds Gemini04:10 Eleven years at Google, from Search and YouTube to leading Gemini06:20 Why the model team works like an engine maker for a garage of cars08:40 Why stack-ranking every request would produce "a boring product"11:25 "Nothing can replace a great eval": evals as the new PRD14:36 The "20 prompts" trick for turning a vague ask into a great eval18:44 How to know a model is ready to ship without waiting a year22:40 "Vibes": why Simon Tokumine's gut check on a model carries real weight27:20 Why speed beats trying to build the perfect product today29:35 Prototyping features that "don't quite work yet," on purpose31:05 The 60/20/20 rule for building against a model that keeps improving33:58 The five traits she looks for in a research PM40:35 Why "technical" on her team means thorough, not a CS degree44:25 How AI tools are changing what Google's PMs do all day48:30 Why Google's I/O still matters as a forcing function52:50 The vibe inside Gemini right now: hype and chaos at once55:20 What she'd change running product outside a model lab57:15 The Nano Banana moment that "fuels all of the other chaos"–Brought to you by:• Chargebee—Billing and monetization infrastructure for the AI era–Referenced:• Antigravity• Gemini• Gemini Deep Research• Gemini Notebook• Gmail• Google• Google Cloud• Google's Associate Product Manager (APM) Program• Inside PM at Meta• Inside PM at Meta 2• Inside PM at Stripe• Inside PM at Waymo• Nano Banana• Simon Tokumine–Where to find Nikhyl:• Twitter/X• LinkedIn–Where to find Tulsee:• Twitter/X• LinkedIn–Join The Skip:• Skip Coach• Skip Community–Find The Skip:• Website• Substack• YouTube• <a target="_blank" href="https://open.spotify.com/show/4oT2rxMsRP