Sara Hooker

Sara Hooker grew up in Africa — Mozambique, Eswatini, and beyond — before a research career at Google Brain, where she worked on model efficiency and interpretability and founded the nonprofit Delta Analytics. Her 2020 essay “The Hardware Lottery” became one of the field’s most-cited critiques, arguing that research ideas win or lose based on how well they fit existing hardware rather than on merit.

Betting against the scaling race

From 2022 to August 2025 Hooker built and led Cohere Labs (originally Cohere For AI), the research arm of the enterprise-AI company founded by Aidan Gomez, where she drove the multilingual Aya model project spanning over a hundred languages. She left to co-found Adaption Labs with fellow Cohere veteran Sudip Roy, raising a $50 million seed round to build what she calls “thinking machines that adapt and continuously learn” — smaller systems that learn from their environment cheaply, instead of ever-larger frozen models.

Hooker’s contrarianism is directed at the scaling orthodoxy, not at AI itself: her complaint is that a handful of frontier labs serve identical, expensive-to-adapt models to everyone, and her answer is to build better ones. If adaptive learning works, she argues, it will democratize who gets to shape the technology — a builder’s critique, prosecuted by founding a company.