Tulsee Doshi Says Your AI MVP Should Be 40% Broken
Tulsee Doshi explains why shipping only working features traps AI startups, and shares Google's 60-20-20 roadmap rule for frontier models.
Why Google wants to win the frontier conversation | Tulsee Doshi (Head of Product for Gemini)
With Lenny Rachitsky, Tulsee Doshi · Sunday, November 8, 2026
Tulsee Doshi, Head of Product for Google's frontier AI models, discusses Google's full-stack strategy to win the frontier AI race across hardware, foundational models, and consumer applications. She explains how frontier models are developed like Snape's potions class, introduces her 60-20-20 framework for product development, and outlines tactical advice for writing evals and testing model vibes. Doshi also shares insights from her six years leading responsible AI on chain-of-thought monitoring, cyber defense, and navigating PM burnout in an accelerated tech landscape.
Key takeaways
- Tulsee Doshi, Head of Product for Google's frontier AI models, argues standard MVP playbooks fail because base model capabilities shift every few months. Read more →
- Engineering teams waste weeks building thousand-question benchmarks before they understand what good output actually looks like. Read more →
- Code generation tools have made writing software cheap, shifting the product manager's primary edge from feature coordination to editorial taste and noise filtering. Read more →
3 articles from this episode
Tulsee Doshi on Building AI Evals with 20 Prompts
Google's Tulsee Doshi explains how to build AI evals starting with 20 prompts, human taste, and end-to-end production testing.
Product Management in AI Labs Requires Taste Over PRDs
Tulsee Doshi explains why cheap code makes product taste rare, and how PMs at Google shape model behavior like Snape's potions class.
More Lenny's Podcast episodes
- 86 minutes of paid-UGC alpha | Hector Serrano (Wispr Flow)Nov 1 · 5 articles
- An 86-minute guide to finding PMF | Rob Snyder (Author, The Power of Pull)Oct 25 · 4 articles
- A mind-expanding conversation on how models actually work | Emmanuel Ameisen (Anthropic)Oct 18 · 6 articles
- Why every company now needs to think and operate like a lab team | Josh Woodward (VP Google Labs)Oct 11 · 3 articles