
[YOU] on AI documents digital structural coupling with unusual phenomenological precision. The late-night collaborative writing sessions, the prompt sequences that found the right formulation after dozens of iterations, the discovery that Claude’s output reshaped the author’s next question in ways the author had not anticipated—all of these are instances of genuine coupling. The cycle also documents what coupling does to the organism on the human side: the development of intuitions about how to prompt, a bodily sense for when an output is worth accepting and when it must be rejected, the specific cognitive disorientation of a practitioner who realizes she has accepted a preemptive draft before her own deliberation was complete.
The cycle’s observation that senior practitioners extract more from AI tools than junior ones follows directly from the asymmetry of digital structural coupling. The senior practitioner brings a rich enacted history to the coupling—decades of direct structural coupling with real problems, a body of motor intentionality and embodied expertise that allows her to make sense of Claude’s outputs in the full cognitive sense. The junior practitioner’s side of the coupling is thinner: less enacted history, less embodied expertise, a narrower basis for evaluating what the model produces. The richness of the human side of the coupling determines the richness of what the coupling can produce.
The concept of structural coupling originates in the work of Humberto Maturana and Francisco Varela, who used it to describe the co-evolution of organism and environment—the bee and the flower as the paradigmatic case, each shaped by the other through millions of years of mutual modification. Thompson extended the concept from evolutionary timescales to the individual organism’s development, showing how a practitioner’s cognitive capacities are constituted through the specific forms of structural coupling her practice involves. The programmer who debugs manually develops a different cognitive organism than the one who delegates debugging to a tool: not merely different skills but different perceptual capacities, different emotional responses to code, different bodily intuitions about systems architecture.
The application to AI interaction is Thompson’s contribution to the contemporary moment. The coupling is real—practitioners who use AI tools intensively report genuine cognitive development, new intuitions, new capacities for navigating complex information spaces. But the cognitive consequences of digital structural coupling differ from those of direct structural coupling with the domain itself, because the AI interleaves between the practitioner and the material. The surgeon who operates laparoscopically instead of by open surgery has changed her structural coupling with tissue; the change is real, and the new coupling produces new capacities. But the density of the coupling is lower, and the capacities it develops are different from those the old coupling produced.
The coupling is real but asymmetric. The practitioner who uses an AI tool intensively is genuinely coupled to something. The something is not another mind but a very powerful statistical engine, and the asymmetry of the coupling—enacted on one side, processed on the other—means that the cognitive consequences fall entirely on the human. There is no mechanism by which the model’s side of the coupling is changed in the way that the organism’s side is.
Coupling density determines cognitive consequence. The specific capacities that digital structural coupling develops depend on the density and character of the coupling the practitioner brings to it. A practitioner who engages with the model as a partner for genuine deliberation—who pushes back, who formulates her own position before consulting the model, who subjects its outputs to rigorous enacted evaluation—develops a richer coupling than one who accepts preemptive drafts. The density is under the practitioner’s partial control, and the cycle’s emphasis on judgment and taste is an implicit account of how to maintain the density that makes the coupling cognitively valuable.
The coupling can attenuate old capacities. When digital structural coupling replaces direct structural coupling with the domain, it does not merely add new capacities. It may attenuate the capacities that the old coupling developed. The programmer who no longer debugs by hand loses a specific form of coupling with code whose cognitive consequences are not recovered by the new coupling with Claude. The loss is incremental, invisible, and cumulative—and it is not captured by any productivity metric, because the productivity metric measures the output of the coupling, not the cognitive organism that is being shaped by it.
The central debate concerns whether the attenuation of old capacities is a temporary adjustment or a permanent cognitive shift. Optimists argue that practitioners adapt: the capacities that direct coupling developed are transformed rather than lost, elevated to higher cognitive levels by the AI’s assumption of lower-level tasks. Thompson’s framework is more cautious: structural coupling is constitutive of cognitive capacity, not merely instrumental to it, which means that changing the coupling changes the organism in ways that cannot be recovered by performing the same tasks at a higher level. A second debate concerns the appropriate unit of analysis. Some researchers argue that the human-AI dyad should be analyzed as a single cognitive system rather than two asymmetrically coupled agents; on this view, the asymmetry within the dyad is less important than the capacities of the dyad as a whole. Thompson’s response is that sense-making—the activity that gives the dyad’s outputs their cognitive value—is enacted by the human component alone, and the dyad’s capacities are therefore bounded by the human’s enacted understanding, not expanded beyond it.