Sand vs Stone — if models double in capability tomorrow, what washes away and what remains?

Barry McCardel’s framework for prioritizing AI product investment: ask “what’s sand and what is stone? If the models get twice as good tomorrow, what turns out to have been sand that can be washed away and what is stone that remains?” Izzy Miller’s answer at Hex: UX polish and domain opinions are a small slice of the pie, but the context flywheel — accumulated artifacts, verified knowledge, feedback loops — is the majority of durable value.

This is the operational version of Context is the product, not the model — turned from a principle into a prioritization tool. Every feature investment gets filtered through: will this survive a model upgrade? The framework explains why Scaffolding is tech debt against the next model — the bitter lesson applied to product building — scaffolding is definitionally sand. And it sharpens Domain-specific skill libraries are the real agent moat, not core infrastructure — domain workflows are stone precisely because no model upgrade can generate years of accumulated domain expertise. The practical implication for product builders: before investing engineering time, classify the work as sand or stone. Sand is acceptable when it’s cheap and temporary, but the strategic investment should flow to stone.