The Decoherence Trap: AI slop and the coming Decoherence Collapse

The essay frames decoherence as a familiar kind of technical debt, though it argues it is the worst form. It uses the example of a building's pipes to illustrate how replacing them piecemeal creates a patchwork that is more complex than either system alone. When changes are needed — like routing water to an annex and increasing pressure — the mixed system demands far more analysis. The author notes that management pressure to move faster leads engineers to skip steps, making the system more fragile and complex, which in turn makes the next change even harder. This creates a cycle of exponential growth.

The essay then addresses whether AI agents can avoid this trap. It cites studies suggesting more code is written without more finished work, with extra code piling up in review. It argues decoherence grows with change, not time, and that AI-driven changes could accelerate it. Finally, the author pushes back on the idea that AI will recursively self-improve, arguing that decoherence both prevents AI from improving in a bubble and limits its improvement outside one. The essay calls the alternative a "Decoherence Collapse," a different kind of disaster than the intelligence explosion.