Sharebird
Sign Up / Sign In
Puja Hait

AMA: Google Group Product Manager, Puja Hait on AI Product Development

December 3, 2025 @ 9:00PM PT

View AMA Answers

  1. How do you balance shipping fast to stay competitive versus ensuring AI features are reliable and safe?

    Puja Hait
    Puja Hait

    Google Group Product Manager • 6mo

    The rule of thumb is that User Safety and Privacy trump everything else. This is further accentuated in Enterprise usecases.

    Reliability is important but you may argue that when I search for sneakers, if I get slightly different but good results, user trust is not impacted. But if the feature or domain is completely falling behind competition, then quick iteration with a Labs like release (rather than general availability) is a less risky way to test the waters and continue velocity/innovation.

    3,086 Views
    1 request

  2. How do you adapt the development process for ML/AI features where data and model performance introduce uncertainty?

    Puja Hait
    Puja Hait

    Google Group Product Manager • 6mo

    In GenAI product development- Your role as a PM essentially shifts from writing a PRD to writing Evals in addition to PRD. The Evals are your reality check on how well the model is actually performing for your intended usecases / user queries. Also the Eval metrics matter a lot. I would treat them as input metrics and engagement (e.g DAU) as output metrics. Its because if your output quality is great, users will come back so focus on making your solutions really shine and matter. Ofcourse you st ...Read More

    2,234 Views
    1 request

  3. How do you decide which problems are truly worth solving with AI versus traditional software approaches?

    Puja Hait
    Puja Hait

    Google Group Product Manager • 6mo

    My litmus test are the following: 1) Can AI do this better - not just the first time but over time? If yes, then its worth the initial investment of training for quality, safety, guardrails etc. 2) Is there lack of data and more judgement required? if so, then AI is an assistance and not a replacement. 3) Does the solution need to deterministic or probabilistic is okay? Perhaps Personalized is better and hence probabilistic? Consumer expectations are changing and what we considered Traditional s ...Read More

    1,819 Views
    1 request

  4. How do you gather meaningful user feedback on AI features that might fail in unexpected ways?

    Puja Hait
    Puja Hait

    Google Group Product Manager • 6mo

    With GenAI, user feedback comes in the traditional forms as well as Thumbs up, Thumbs down Query edit and retries. Users are prompt engineering constantly. The perception is that users are not "prompting well enough". But as a PM, your job is to make the users life easier, effective, better. So understanding the intent of especially multi-turn queries is important to effectively measure quality. A well reflecting Quality rubric is more important than DAU/WAU/MAU number. e.g Factuality, Instructi ...Read More

    839 Views
    1 request

  5. What's the biggest misconception about building AI products that you've had to navigate?

    Puja Hait
    Puja Hait

    Google Group Product Manager • 6mo

    In my experiences, the biggest challenge is to change user behaviors fundamentally. Its far easier to insert AI into something users already do versus create a net new surface or user journey for them.

    Also its not always straightforward to paint what surface should be used for what- since with GenAI, you can do a lot on the same surface so lines start to blur. Its important to build and iterate with an open mind since nobody holds the crystal ball.

    834 Views
    1 request

  6. What does your cross-functional team structure look like for AI products? How is it different from traditional product teams?

    Puja Hait
    Puja Hait

    Google Group Product Manager • 6mo

    I see a few changes Evals is a new area. This is led by PM or Program Manager in conjunction with Data Science, User Research and often involves vendors for scaling. Trust & Safety evals in addition to Quality checks become an important part of cross functional team set up. Between engineering, product, design there is a harmony of understanding where the model and product are lacking and what areas need to be solved by model vs product or design. e.g File picker in Docs Gemini Sidepanel sol ...Read More

    836 Views
    1 request