Accidental Scaffolding

In the [YOU] on AI Field Guide

The cycle’s treatment of the AI productivity transition is balanced between celebration and diagnostic honesty. The celebration is warranted: the elimination of accidental complexity at the scale AI achieves is a genuine liberation, freeing builders to work at the level of intention rather than the level of syntax. The diagnostic honesty requires naming what the liberation costs. Accidental scaffolding is that cost, stated precisely: not the loss of the tedious work itself, which was always waste, but the loss of the understanding that the proximity to tedious work accidentally produced.

The concept connects directly to the cycle’s account of the solo builder’s deepest vulnerability. The solo builder can produce what a team previously produced—but she cannot produce the accumulated intuition that the team’s collective years of proximity to the problem deposited in each of its members. The team’s communication overhead was a cost; it was also a structure within which diverse proximity to the same problem produced diverse intuitions that collided productively. When the AI absorbs both the execution and the proximity, the solo builder must find other means of developing the understanding that proximity previously delivered. The cycle treats this as the central skill-development challenge of the AI age: not learning the tools, but learning how to accumulate the essential understanding that the accidental complexity used to force on the builder inadvertently.

Origin

The concept emerges from the juxtaposition of two observations in Brooks’s work. The first, from “No Silver Bullet,” is that essential and accidental complexity are separable: better tools can eliminate accidental complexity without touching the essential. The second, from The Mythical Man-Month’s chapters on the surgical team and project planning, is that developers who have worked through difficult implementations develop a feel for systems that transcends any individual technique—an architectural instinct built through thousands of encounters with systems that behaved unexpectedly.

The AI transition has produced the empirical evidence that these two observations are in tension: the developers who used AI tools extensively and early report, with some regularity, a degradation of exactly the architectural intuition that Brooks identified as the hard-won product of implementation experience. The tool that eliminates the difficulty also eliminates the experience-through-difficulty that the difficulty was generating. Neither Brooks nor his contemporaries could have fully anticipated this dynamic, because the tool capable of producing it did not exist until 2023. The dynamic is, however, fully predictable from the framework he built, and the observation that it has occurred is a confirmation of his deepest insight: that essential complexity and accidental complexity are not as separable as they appear.

Key Ideas

Proximity as the mechanism of understanding. Essential understanding of a software system—the intuitive sense of how its components relate, where its stress points are, which changes are safe and which are dangerous—is not acquired through documentation or specification but through proximity: the sustained, repeated, often involuntary encounter with the system’s actual behavior under conditions that differ from the expected. Four hours debugging a data access layer is not four hours of productive work; it is four hours of involuntary proximity to the data model, during which a thin layer of intuition is deposited that the developer cannot articulate and did not seek. Accidental complexity provided this proximity; eliminating it eliminates the mechanism.

The scaffold is not the building. The accidental complexity that AI eliminates was always overhead—always waste, never the thing the developer was trying to build. The scaffold metaphor is exact: scaffolding is not the building, and removing it before the building is structurally complete leaves the builder in open air. The point is not to preserve the waste but to find other means of achieving the proximity that the waste accidentally enabled—means that are deliberate rather than inadvertent, and therefore available to the builder who understands what she is seeking.

The warning system silenced. Before AI, the tar pit announced itself gradually: implementations became harder, debugging took longer, integration produced unexpected failures. The increasing resistance of the tools signaled that the essential complexity of the problem had increased, giving the builder time to reconsider before the commitment became irrevocable. AI generates code at the same speed regardless of the essential complexity of the direction. The warning that came from the resistance of the tools is silent; the builder does not feel herself sinking until she is already deep, and the system she built rapidly and confidently fails in ways that reveal a mismatch between her understanding of the problem and the problem itself.