Skip to content

By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

Lenny's PodcastLenny's Podcast

Why Google wants to win the frontier conversation | Tulsee Doshi (Head of Product for Gemini)

Tulsee Doshi is head of product for Google’s frontier AI models, including Gemini and other multimodal models, such as Nano Banana, Genie, Veo, and Lyria.

Tulsee DoshiguestLenny Rachitskyhost
Nov 8, 202658mWatch on YouTube ↗

Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.

At a glance

WHAT IT’S REALLY ABOUT

How Google’s Gemini PM thinks about frontier models and product

  1. Tulsee Doshi explains that even with clear goals, frontier model training remains partly unpredictable, so PMs must understand a model’s “vibes” and productize the surprising strengths that emerge.
  2. She describes Google’s AI strategy as full-stack and portfolio-based, aiming to compete at the frontier while also offering models across the cost–quality curve for many different user types and use cases.
  3. On AI safety, she emphasizes embedded safety work from training through deployment, highlighting monitoring and intervention mechanisms and also pointing to AI’s role in strengthening cyber defenses.
  4. She argues that UX includes both the product interface and the model’s behavior/persona, and that PMs can translate desired behavior into specs, rubrics, evals, and ultimately training signals.
  5. She shares practical PM craft for the current AI era: build small evals first, develop judgment through real use cases, invest in upskilling via shared workshops and reusable internal “skills,” and build products with the 60-20-20 rule so they evolve with model improvements.

IDEAS WORTH REMEMBERING

5 ideas

Frontier models are designed with goals, but they still behave unpredictably—PMs must productize the surprises.

She says model development should be intentional about goals and evals, but outcomes still include “magic” that can’t be fully predicted; a PM’s job is to notice what unexpectedly shines and shape it into user value.

Google’s AI strategy is to compete at the frontier while serving many user segments across cost vs. quality.

She describes Google’s approach as “full stack,” spanning hardware through products used at massive scale, and emphasizes building a range of models across the cost–quality spectrum for different user needs (from Search to engineering to cyber defense).

Safety isn’t just policy—deployment monitoring and intervention are central mitigations.

Drawing on her responsible AI background, she highlights safety being embedded from training through deployment, and points to monitoring as a key mitigation—specifically monitors that assess whether model behavior aligns with directed actions and enable intervention.

The model’s behavior and persona function as UX, and PMs can specify and measure it.

She argues UX isn’t only interface design; the model itself has UX characteristics like transparency, communication style, and what it asks users, and PMs can help translate “good behavior” into rubrics and reward-modeling inputs.

As building gets cheaper and faster, PM value shifts to taste, insight, and alignment.

She says PMs are becoming more valuable because when building is easy, “taste” and insight matter more, and because speed without coordination can create chaos—PMs help turn chaos into order and align teams.

WORDS WORTH SAVING

5 quotes

The model isn't predictable. It's kind of like Snape's potions class. You can put all the ingredients in, but you also have to expect that there will be a little bit of magic, and that magic is going to be surprising.

— Tulsee Doshi

I actually think we do want to win the frontier conversation, right?

— Tulsee Doshi

It's also important to think about the model as having a UX also, right?

— Tulsee Doshi

When you can build anything, and you can build anything so easily, building something great actually is now a, a unique value add, right? Um, and taste is now really important.

— Tulsee Doshi

Burnout comes not just from working a lot. It comes from, I think, two places. I think it comes, one, from when you feel like there's churn and chaos. I think that introduces burnout. I think the other thing that introduces burnout is work that you don't enjoy.

— Tulsee Doshi

Read the available transcript
Model unpredictability and “magic” in trainingGoogle full-stack AI strategy and cost–quality tradeoffsSafety mitigations: monitoring, alignment, interventionAI for cyber defense and vulnerability patchingModel behavior as UX (persona, transparency)Evals vs. vibes; building small evals60-20-20 product-building framework and harness adaptability

SPEAKERS

  • Tulsee Doshi

    guest

    Head of Product for Google’s frontier models (Gemini), with experience in responsible AI and model evaluation/behavior.

  • Lenny Rachitsky

    host

    Host of Lenny’s Podcast and interviewer focused on product management and technology.

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.

iOSAndroidClaudeChatGPTChrome