The Marshmallow Was Never the Hard Part

There's an experiment a designer named Tom Wujec ran for years, in dozens of countries, across nearly every professional category: teams of four get twenty sticks of spaghetti, some tape, some string, and one marshmallow. Eighteen minutes. Build the tallest freestanding structure, marshmallow on top.

He ran it with engineers, architects, CEOs, and graduates of the most competitive business schools in the world. Then, almost as an afterthought, he ran it with a room full of kindergarteners.

The kindergarteners won. Not "did well for their age" — won, on almost every measure. The business grads spent most of their eighteen minutes debating roles and planning an approach, which left them no time to recover when the marshmallow — always assumed to be the easy part — collapsed the whole structure at the last second. The kindergarteners never spent a minute deciding who was in charge. They just started building, failed, and built again.

I think about that story reading the numbers on AI adoption right now: 80% of executives report using AI regularly. Only 20% of frontline employees do. And in the organizations where frontline teams are actually given room to experiment together — not handed a tool and a mandate, but allowed to build, fail, and try again — the rate of real AI-driven improvement is roughly double.

The marshmallow was never the hard part. Neither is the AI. The hard part is whether whoever's holding it is actually allowed to just start building.

Who in your organization is closest to the marshmallow — and are they allowed to touch it yet?

From Dancing in the Wind: An Adaptive Leadership Framework in the Age of AI.