Inside Gemini: How Google Runs Product for Its Model

Product management in frontier AI labs requires a fundamental shift from traditional metrics-based evaluation to deep, hands-on model engagement. Success hinges on building "mini evals"—small, replicable sets of prompts that define what "good" looks like—to guide researchers and identify model losses. Rather than relying solely on quantitative data, effective PMs must develop an intuitive understanding of a model’s personality and capabilities through constant prototyping and demo building. This approach allows teams to identify glimmers of potential in emerging technology before it is fully realized. Prioritization involves balancing first-party product surfaces with broader research goals, using iterative shipping to gather real-world feedback. Ultimately, the role demands high intellectual curiosity and the ability to bridge the gap between technical research and user-centric product design in an environment of rapid, continuous innovation.