Model-Continuous Experience

In the AI Story

Model-continuous experience is not valueless. The builder learns genuine things: how specification relates to implementation, how descriptions can be structured for better interpretation, how the tool succeeds and fails across different kinds of problems. These are transferable insights about the interface between human intention and machine interpretation. They have durability in the sense that the general shape of prompt engineering survives particular model updates.

But the knowledge is about the tool, not the domain. A builder who has spent two years becoming expert at Claude Code has developed significant model-continuous understanding. If the underlying architecture changes, or if a competitor's tool with different interpretive patterns becomes standard, she faces substantial rebuilding. The experienced surgeon whose domain-continuous knowledge of anatomy transfers across instruments does not face this problem. Her tools change; her domain does not.

The deepest concern is not that model-continuous experience is inferior to domain-continuous experience in every case. It is that model-continuous experience may be mistaken for domain-continuous experience by the builder herself. The working product creates the impression of understanding where understanding of the domain does not yet exist. The novice who produces a functional application through AI-mediated description may believe she understands software because the evidence — the working artifact — supports the belief. Only future experience reveals the gap: when the model changes, when a novel problem arises, when the accumulated tool-understanding proves inadequate to the situation.

This makes AI-augmented building most educative for those who need it least. The expert with twenty years of domain-continuous experience uses AI as an amplifier of existing understanding and can distinguish what the tool produces from what the domain requires. The novice without that foundation uses AI as a substitute for understanding she has not yet built. The tool does not advantage the experienced out of malice; it advantages them because they possess the interpretive framework that transforms model-mediated experience into domain-relevant learning.

Origin

The term was coined in the Dewey volume as a diagnostic complement to domain-continuous experience, both operationalizing Dewey's principle of continuity for the AI era. The framework extends Dewey's distinction between principled and procedural knowledge, giving it contemporary relevance for tool-mediated practice.

Key Ideas

Tool-keyed knowledge. The understanding deposited is about the model's interpretive patterns, not the domain's enduring logic.

Fragile transferability. Model updates can invalidate substantial portions of the accumulated understanding.

The indistinguishability problem. Working products are produced by both domain-continuous and model-continuous experience; only the builder's future encounters reveal which kind of understanding exists.

Asymmetric benefit. The builder with prior domain knowledge transforms model-mediated work into domain-relevant learning; the novice without that foundation cannot.

Debates & Critiques

An emerging counter-argument, sometimes called the convergent expertise view, holds that sufficient model-continuous experience may eventually produce domain understanding indirectly — that prolonged work with sophisticated AI tools teaches the domain through the tool's responses in the way that prolonged work with a piano eventually teaches music. The Dewey volume's framework is skeptical but not dismissive: the question is empirical, and the answer depends on whether the builder engages with the tool's output in ways that force domain-level reflection, or simply accepts the output and moves on.