Tulsee Doshi
Senior Director and Head of Product for Gemini models, Google DeepMind
Stanford and an early interest in AI ethics
Tulsee Doshi holds a BS in symbolic systems and an MS in computer science from Stanford University[1][2]. Profile magazine reports that she spent a quarter studying abroad at Oxford[2]. "I studied the philosophy and ethics of artificial intelligence," she told the magazine[2].
After Stanford she joined Google as an associate product manager[2]. She first met AI ethics in a product context while working on YouTube, and she decided to spend more of her time on the subject[2].
By 2018 she had pitched the leadership of Google's research organisation on having a product manager to drive strategy for responsible AI[2]. Profile magazine describes her as Google's head of product for responsible AI and human-centred technology[2].
Head of product for responsible AI
Doshi spent five years as head of product for responsible AI at Google, building products meant to be more equitable, safe and trustworthy[1]. Forbes reports that she led the launch of more than 30 products to make Google's AI-based work safer and more inclusive[3]. It also reports that she and Engineering VP Marian Croak set up an internal organisation of more than 100 AI researchers and engineers, which she led[3].
In 2022 the magazine reported that, over the previous four years, she had seen the responsible AI team grow from a handful of research scientists and engineers[2]. She told the magazine: "We all have different lived experiences and perspectives that affect the way we think about responsible AI."[2]
TechCrunch's author page notes that she has been recognised as one of the Forbes 30 Under 30 and as one of the top women in AI ethics[1].
Product lead for Gemini
Doshi is now Senior Director and Head of Product for Gemini and generative media models at Google DeepMind[1]. In this role she leads the roadmap and product strategy for Google's foundational language model[1]. Google's blog lists her title as Senior Director, Product Management[4].
In May 2026 she spoke to TechCrunch about the launch of Gemini 3.5 Flash, which Google built around AI agents[5]. She said the model can run on its own for several hours, but it will sometimes pause and ask the user for input at a decision point or permission issue[5].
She described how 3.5 Flash was designed to work with the larger 3.5 Pro model: "3.5 Pro becomes your orchestrator, your planner, and then it actually can leverage Flash to be the various sub-agents."[5] She said the decision comes down to where a team wants the larger model's reasoning power, and where tasks need fast tool use[5].
Gemini 3.8 Flash
On 2 September 2026 Doshi co-wrote Google's announcement of Gemini 3.8 Flash and 3.8 Flash Cyber with Raluca Ada Popa, Gemini Security Lead at Google DeepMind[6]. The post called 3.8 Flash Google's best reasoning and coding model yet, at the same speed and cost as 3.7 Flash[6]. The model was Google's third Flash release in six weeks[6].
According to the post, 3.8 Flash is built for long-horizon coding and autonomous agents[6]. On complex tasks it takes extra reasoning steps and calls tools again and again, and developers can choose lower effort levels to save tokens[6]. The second model, 3.8 Flash Cyber, focuses on finding vulnerabilities and patching them automatically[6]. Google made it available to trusted defenders through a programme called Fairwind[6].
Known for
- Product lead for Google's Gemini models
- Head of product for responsible AI at Google for five years
- Co-author of the Gemini 3.8 Flash announcement (2026)
Sources
- TechCrunch: Tulsee Doshi author page — checked 8 Nov 2026
- Profile: Tulsee Doshi Leads toward a More Responsible Future in AI, 2022 — checked 8 Nov 2026
- Forbes: Tulsee Doshi profile — checked 8 Nov 2026
- Google blog: Tulsee Doshi author page — checked 8 Nov 2026
- TechCrunch: With Gemini 3.5 Flash, Google bets its next AI wave on agents, not chatbots, 19 May 2026 — checked 8 Nov 2026
- Google blog: Introducing Gemini 3.8 Flash and 3.8 Flash Cyber, 2 September 2026 — checked 8 Nov 2026