The Goal: To move Gemini beyond robotic AI responses to a more human-centric, collaborative assistant that engages safely and naturally with users.

The Work: One of my responsibilities is to write Goldens—meticulously scripted narratives that serve as the behavioral DNA for the model. I architect high-stakes interactions where the AI must balance nuance, wit, and empathy. It’s essentially about teaching a machine how to read the room so it can collaborate with a human touch (without embarassing itself).

Additional Responsibilities: Evaluate model responses to create heuristics for collaborativity (it’s a word that Google made up), safety and factuality losses, or “mental short-cuts” that Gemini can use to determine the best and most helpful response in any given scenario. This is foundational to creating systems that Gemini can operate within at scale.

The Skills: Brand Voice, Content Design & Content Strategy. Training a model to speak in a specific voice is like creating editorial guidelines for a brand, then implementing it across platforms. Or like writing a series bible for a character who has to interact with millions of people simultaneously while staying true to its core voice.

The Impact: Gemini has transformed from a “question/answer bot” into a more intuitive, collaborative assistant with a voice and personality that keeps users engaged. Next up: Training Gemini to not just speak properly, but to also do things correctly (agentics).