The author reflects on a decision on whether to prioritize building basic features or a learning infrastructure first for a software project.
They opt to build learning scaffolding first to capture signals from the beginning, emphasizing the importance of starting the learning process early.
They describe the development of a minimal learning infrastructure as 'plumbing' that sets a foundation for future learning algorithms and features.
The approach contrasts with traditional methods by focusing on building learning capabilities before adding features, with the belief that the infrastructure created will pave the way for successful integration of advanced AI in the future.