Why PMs are suddenly valuable again | Tulsee Doshi (Head of Product, Gemini)
Lenny's Podcast · 58:33 · transcribed 5d agoTech
TL;DR Product Managers are valuable again because they provide vision and structure in a chaotic, unpredictable AI landscape. Google aims to lead the industry by building diverse models that serve users across all cost and quality levels.
Product Managers must understand model behavior to guide unpredictable technology.
- Models are not fully predictable like traditional software products.
- Managers must test models on topics they care about deeply.
- This testing helps identify unique strengths and unexpected capabilities.
- Managers shape the final product based on how the model actually behaves.
Google builds a wide range of models for different user needs.
- The strategy covers hardware, core models, and end-user applications.
- Goals include serving researchers, engineers, and security professionals simultaneously.
- Products aim to balance cost and quality for various use cases.
- The company seeks to win leadership while improving daily life for billions.
Safety relies on monitoring and using AI to secure systems.
- Teams monitor model thought processes to detect unsafe behavior.
- Interventions occur when models deviate from their intended instructions.
- AI partners help identify vulnerabilities in global infrastructure.
- Continuous dialogue about safety reduces risks and builds confidence.
New product frameworks allow building features before they fully work.
- The 60-20-20 rule guides feature development planning.
- Sixty percent of features should function reliably today.
- Twenty percent may be imperfect but show clear potential.
- The final twenty percent targets future magical experiences as models improve.
Hiring focuses on intuition, technical skills, and adaptability.
- Recruiters seek managers who understand the "character" of models.
- Candidates must analyze data and build evaluation metrics themselves.
- Flexibility is essential due to rapid weekly changes in the market.
- Managers must constantly rethink strategies rather than relying on fixed plans.
Next steps
- Test AI models on personal projects to build intuition.
- Start evaluations with small sets of prompts to find patterns.
- Use AI tools to automate repetitive tasks and reduce burnout.
- Provide feedback on AI products to help developers improve them.