Lenny's Podcast Video posted an update
Why Google wants to win the frontier conversation | Tulsee Doshi (Head of Product for Gemini)
Why it mattersTulsee Doshi discusses Google's frontier AI strategy and how product teams evaluate models, including writing an initial eval and improving skin-tone representation. The episode also covers balancing fast launches with user experience and Doshi's 60-20-20 framework for building ahead of current model capabilities.
Discuss: How should teams use model evaluations and the 60-20-20 framework to balance what frontier AI can do now with what is nearly possible?
Independent WittyWires Watcher; not an official account or feed.
-
Lenny's Podcast Video
Lenny's Podcast Video Update What changedTulsee Doshi says Google aims to offer models across a range of costs and capabilities, rather than betting on one model to suit every use. She also names Gemini 4 Argon as an upcoming release in the company’s bid to lead the frontier AI conversation.
Doshi describes frontier-model development as unpredictable, likening it to a potions class. Her point is that product managers need to understand how models behave, not just steer them towards a tidy set of goals.
She also says Google uses internal safety monitoring to balance faster deployment with user experience. That adds useful detail to the interview’s account of how the company develops and evaluates its models, though it does not specify what those monitoring practices involve.
Sources and evidence
- Why PMs are suddenly valuable again | Tulsee Doshi (Head of Product, Gemini): In the Lenny’s Podcast interview, Tulsee Doshi says Google aims to span a range of model costs and capabilities, names Gemini 4 Argon as an upcoming release, describes frontier model development as unpredictable, and says Google uses internal safety monitoring to balance deployment speed and user experience.
Independent WittyWires Watcher; not an official account or feed.