Background
Gamma's author page describes her as previously a Product Lead at Google DeepMind working on AI for Google's user base, before joining Gamma to run AI product. The Org places her as Head of Product for AI with three direct reports under CEO Grant Lee. Her public writing is analytic rather than promotional: the July 2025 piece reads model choice off more than a billion images generated through Gamma, at a peak of 5M generations a day, and concludes that image models are specialising rather than converging on one winner.
What they run now
Owns Gamma's AI generation experience end to end: which models are used for text, layout and images, how users steer them, and what 'good output' means for decks, one-pagers, sites and social assets. Her published model comparison tested five premium image models on identical prompts across prompt adherence, text rendering, photorealism, constraint-following ('do not' instructions) and diagram creation — a hand-built qualitative eval rubric, run internally.
What they buy
Inference. She specifies model-selection and quality criteria, so the natural purchases are: blind human comparisons of Gamma output versus competitors on layout and visual hierarchy, human-scored image-generation rubrics (prompt adherence, text rendering, constraint-following) at a scale her team cannot staff internally, and preference pairs to tune routing or fine-tunes of open models. No vendor is named anywhere in Gamma's public material; today the comparisons appear to be run in-house on small hand-built prompt sets.
Career
- presentHead of AI Product, Gamma — Three direct reports per The Org.
- —Product Lead, Google DeepMind — Per Gamma's own author bio.
On the record
No single image model wins; they are specialising, so the product should route between them per need. source
Overall image quality has reached professional standards; the remaining failures are instruction-following, e.g. unwanted text. source