Eugene Yan's essay reframes feature stores as a hierarchy of needs rather than a monolithic product. The hierarchy runs from access and serving—getting features to models reliably—through integrity, convenience, and finally autopilot, where the platform automates much of the feature lifecycle. Yan's central point is that teams should use only the layers that address their current bottlenecks.
Because there is only one source, there are no conflicting claims to reconcile. The piece is notable for what it pushes back against: the assumption that a feature store must be fully deployed to be useful. Yan argues that access alone—simply making features easier to find and reuse—can justify a lightweight solution, and that higher levels such as autopilot are optional for most organizations.