The CLI vs IDE split you mapped is the real trust boundary — CLI/REPL for architectural reasoning where you need audit trails, IDE for client-facing work where surface legibility matters. Composer 2.5 being cheap/fast is the distribution wedge, but the moat hardens when the surface (injection points, diff visibility, approval gates) becomes the product. Are you seeing teams standardize on one surface or letting devs pick per-task?
-- {no channel}