Sarah Sachs describes the economics of building AI products when model suppliers also compete with their customers. Drawing on Notion's experience, she argues for understanding complete task trajectories rather than token prices alone, matching model capabilities to the work and preserving the ability to change providers.
Sarah Sachs discusses open-weight alternatives, deterministic CPU-based workers and governance as ways to deliver useful AI without paying for unnecessary model calls. She then examines agent security, persistent enterprise context and a collaborative task workflow in Notion, presenting product demonstrations and efficiency claims as Notion's account rather than independently verified results.
Watch on YouTube
