Friction

In the AI Story

The parasitic friction is the boilerplate, the dependency hell, the configuration fiddling, the mechanical labor that consumes time without building capability. Its removal is genuine gain. The developer freed from managing package.json conflicts can direct her attention to problems that actually require her judgment. This is liberation in the unambiguous sense. Leopold would have called it sensible — the removal of waste that served no function.

The regulatory friction is different. The debugging session that produces no useful code but deposits a layer of understanding about how systems fail. The research expedition that dead-ends but teaches the researcher what the territory looks like. The collaboration that generates disagreement before consensus, building shared understanding in the disagreement. These processes look wasteful from the sprint perspective. They look like the wolf looked to the rancher: costs to be eliminated. They are the regulators. They maintain the cognitive ecosystem's capacity to sustain itself.

The AI discourse typically treats friction as a single category — either all good or all bad. The triumphalist asserts that friction is waste and its elimination is pure gain. The Luddite asserts that friction is sacred and its elimination is pure loss. Both positions substitute conviction for observation. The ecological perspective demands the harder work: distinguishing, in specific cases, which friction is wolf and which is parasite. The distinction cannot be made in general. It requires attention to the specific system, the specific practitioner, the specific stage of development.

The practical implication is that the same AI tool can be a liberator or an erosion depending on how it is used and by whom. The senior engineer whose understanding is already deposited can use AI to skip parasitic friction without loss — her capacity to read the system is already built. The junior engineer whose understanding is still forming uses the same tool and skips the regulatory friction that would have built her capacity. Same tool, opposite effects. The difference is not in the tool. It is in the relationship between the tool and the user's developmental stage.

Origin

The concept of friction in cognitive and organizational contexts draws on work from Karl Weick, Eric Trist, and others in the sociotechnical systems tradition. The specific framing here combines that tradition with Leopold's ecological categories.

Key Ideas

Friction is not one thing. It is a bundle of processes. Some parasitic, some regulatory.

Same tool, different effects. The same AI interaction supports one practitioner and erodes another, depending on developmental stage.

The triumphalist and the Luddite are both wrong. Neither the 'all friction is waste' position nor the 'all friction is sacred' position survives contact with specific cases.

Distinguishing requires attention. No formula resolves the question. Each case is specific. Ecological literacy is the meta-skill.