Innovation cannot be compelled. No transformation office, quarterly OKR, or CEO all-hands deck has ever produced it. Something novel and useful — which is all innovation means, stripped of mysticism — emerges only when the enabling conditions exist. Leadership sets those conditions. Everything else is theater.

This distinction between compelling and enabling carries more consequence than it appears to, because it determines how an organization relates to AI at the structural level. Most enterprises treat AI as a product — a thing to procure, configure, and deploy against a use case. But AI belongs to a different category entirely: it operates as a general-purpose enabling technology (GPET), the same class as electricity, the internal combustion engine, and the transistor. Only a handful of technologies in the last two centuries qualify. What they share is a thermodynamic signature: they arrive looking like tools, persist as infrastructure, and dissolve the constraints that shaped the existing economy. The economy then reassembles, over decades, around new degrees of freedom that were literally unthinkable under the prior physics.

Electricity did not automate candles. It decoupled manufacturing output from daylight hours, factory location from waterway proximity, building height from human-climbable staircases. Each dissolved constraint permitted structures that no one designed in advance, because no one could imagine them while the constraint still held. That is what emergence looks like in economic systems: you set the conditions, and the structures self-organize. You cannot specify the output. This is why command-and-control fails with innovation — and why $547 billion of the $684 billion enterprises invested in AI in 2025 failed to deliver intended business value. The technology worked. The organizational conditions did not.