Prompting is a velocity input

Prompting tells the computer how to move; direct manipulation tells it exactly where to stop

In 1978, Card, English, and Burr at Xerox PARC showed that direct pointing with a mouse outperformed the alternatives they tested in both speed and accuracy because users could point directly to a target in one movement. This result became one of the foundations for modern graphical interfaces.

Less discussed are the other devices tested. While the mouse pointed at a target, the joystick expressed a velocity: keep moving in this direction. Although effective for exploration, joysticks required iterative refinement to reach a precise destination.

Prompting is AI's velocity input. Like a joystick, it communicates how a system should move, rather than where it should end up. It launches a search through an enormous solution space with remarkable efficiency, making it an incredible interface for exploration, ideation, and rapidly converging on promising directions.

But exploration eventually gives way to refinement, and every constraint comes into focus. Direct interaction with the artifact—be it code or design—provides a fidelity that is difficult to capture through natural language alone. Every dependency, every pixel, every interaction matters.

Each interaction model excels at a different level of precision. In a nutshell, prompting tells the computer how to move, while direct manipulation tells it exactly where to stop.