The Compression Curve

In the AI Story

The adjacency principle is essential to the Miller-framework analysis of compression. Recoding works by transforming existing chunks into new, higher-level chunks. The existing chunks must already be present — built through prior experience — before they can serve as raw material for the next level. A developer who has built chunking vocabularies for database design, API architecture, and frontend rendering can absorb a framework like Rails relatively quickly because the framework's abstractions map onto chunks she already possesses. The framework compresses what she already understands.

The current AI compression violates the adjacency principle. Claude Code does not compress a single adjacent layer of software development. It compresses multiple layers simultaneously. A developer using such tools does not need to have mastered database design to produce software that interacts with databases. The tool handles implementation across the entire stack. The compression is not one level above the developer's current expertise but potentially five or six levels above, skipping intermediate stages where recoding would traditionally have occurred.

The calculator offers the closest historical precedent. Students who used calculators early could engage with higher-level mathematical concepts sooner but sometimes found themselves stranded when higher-level work required an intuitive grasp of numerical relationships the calculator had prevented them from building. A calculus student who has never performed long division by hand may lack the embodied sense of what division means. The calculator did not eliminate the need for arithmetic understanding; it made such understanding optional for producing correct answers while leaving it essential for producing deep understanding. The parallel to AI-assisted development is precise.

The compression curve has now reached a point where its speed exceeds the speed at which recoding infrastructure can adapt. This is not a temporary mismatch that educational systems will resolve. It is a structural consequence of a compression that eliminates the very experiences through which recoding occurs. You cannot build recoding infrastructure for implementation skills when the tool has made implementation invisible. The medium through which recoding would happen has been compressed out of existence.

Origin

The observation that technology adoption curves have steepened over time is widely documented in innovation studies, though few frameworks integrate this steepening with cognitive-science constraints. Everett Rogers's diffusion of innovations research established the empirical S-curves; adoption curve compression across successive generations of technology has been a recurrent theme in technology history.

The synthesis presented here — connecting the compression curve explicitly to Miller's recoding timeline — draws on Herbert Simon's work on expertise timescales, Ericsson's research on deliberate practice, and contemporary observations about generational differences in AI-mediated work.

Key Ideas

Intervals shrinking from centuries to months. Writing's implications took a millennium; printing's a century; AI's capabilities measurably shift within months.

The adjacency principle. Previous compressions were absorbable because each compressed a layer adjacent to the already-understood one. Chunks at the old level provided raw material for chunks at the new.

AI violates adjacency. Current tools compress multiple layers simultaneously, skipping the intermediate stages where recoding would have occurred.

The recoding timescale is biological. Building deep expertise requires approximately ten years of practice. This timescale is rooted in neural learning mechanisms that have not accelerated.

The crisis is structural, not educational. The mismatch between compression speed and recoding speed cannot be resolved by faster schools because the tool eliminates the medium through which recoding occurred.

Debates & Critiques

The debate over whether AI-native practitioners will develop adequate expertise through new forms of recoding — evaluating AI outputs, specifying requirements, orchestrating systems — remains unresolved. One camp holds that the new practice generates its own chunks adequate to the new role. Another holds that these chunks are insufficient because they depend on absent lower-level chunks for their structural integrity. Resolution will come only with time.