When expensive preparation is needed before useful work begins — loading skills and plugins into an AI session, hydrating a dev environment, warming a JIT, fetching a dependency graph, rebuilding a container — the first instinct is *prepare once and clone*. But nearly every layer of the modern stack
When expensive preparation is needed before useful work begins — loading skills and plugins into an AI session, hydrating a dev environment, warming a JIT, fetching a dependency graph, rebuilding a container — the first instinct is *prepare once and clone*. But nearly every layer of the modern stack amortizes setup cost differently — through **prefix caching** — identical inputs reuse the previous