christine mana
Principal designer, AI model behavior
My work focuses on defining what good AI output looks like, and turning it into forms models and teams can use: taxonomies, specs, system prompts, and evaluation criteria. For the past six years I've worked at Meta on AI and wearables, most recently on how Muse, Meta's personal AI agent, behaves on glasses.
Selected work
01 / Model design & evaluations
Teaching an AI how to show and tell
A response taxonomy for Meta AI on Meta Ray-Ban Display, and the ground truth its launch quality was measured against.
02 / Model design
Designing for wait times you can’t predict
A latency strategy for a voice agent on glasses, designed to hold up without knowing how long anything will take.
03 / Evaluations
How should we judge generated UI?
A proposal for a two-tier evaluation framework, written to unblock a decision between two approaches.
About
My path to model behavior ran through language. I started in publishing and communications, then spent years designing content systems: frameworks, information architecture, and taxonomies at SYPartners, and a metadata taxonomy built with data science at Meetup.
At Meta, I spent five years leading content design for the company's first smart glasses and wrist wearables (among other devices), as well as the Wearables program's pivot into AI. I moved back to individual-contributor work in 2025 to work directly on how AI behaves. The skill turned out to be the same one: managing context and structuring meaning so that people, and now models, can act on it.
These days, I'm thinking about how an AI's behavior should change as trust grows (and is sometimes lost) between a person and their agent, and how much AI-generated interfaces should adapt along the way.