Developmental Design

In The You On AI Encyclopedia

The principle connects to Jerome Bruner's concept of scaffolding—temporary support enabling learners to accomplish what exceeds independent capability, designed to be gradually withdrawn as capability develops. The critical distinction: scaffolding builds independent capability; prosthesis replaces it. A well-designed scaffold makes itself obsolete by developing the capacity it temporarily provides. A prosthesis makes itself indispensable by substituting for a capacity the user then does not develop. The language interface operates in the dangerous middle ground—it scaffolds by reducing translation cost and enabling attempts previously impossible, but it also substitutes by producing outputs the user evaluates without having generated, bypassing the struggle through which generative capacity is built.

Developmental design does not require that AI-assisted work be made artificially difficult—that would negate the tools' legitimate value. It requires that the structures of practice include deliberate opportunities for the human to engage at the level of understanding, not just output. For developers: periodic manual coding sessions, not because manual coding is more efficient but because the friction builds understanding making AI-assisted work more valuable. For writers: regular exercises writing without AI, confronting the blank page alone, experiencing the discomfort of not knowing what to say next—the discomfort that is, in Winograd's framework, the condition under which understanding develops. For any practitioner: cycling between tool-augmented flow and tool-free struggle, so the capability expanded by AI is grounded in understanding that can operate independently.

Origin

The principle emerged from Winograd's observation of SHRDLU users in the early 1970s—people who became fluent in describing operations to the blocks world system but did not thereby learn to program. The fluency was domain-specific and tool-dependent. When the tool was removed, the capability vanished. This pattern—apparent mastery that proves to be borrowed rather than owned—became Winograd's paradigmatic concern. His turn to design for intelligence augmentation rather than artificial intelligence was motivated by the recognition that tools claiming to understand would systematically prevent humans from developing the understanding those tools were supposedly providing. The only defense: design the tool and the practice so development is preserved alongside acceleration.

Key Ideas

Scaffolding vs. prosthesis distinction. Temporary support building independent capability vs. permanent replacement preventing capability development—AI collaboration must be structured toward the first, not default to the second.

Understanding through productive friction. The struggle of manual work deposits layers of embodied knowledge; removing struggle accelerates output while preventing deposition unless practice includes deliberate friction-rich engagement.

Cycling between augmented and independent work. Alternating tool-assisted flow with tool-free struggle grounds expanded capability in understanding that can operate when the tool is absent—developmental insurance.

Long-term capability over short-term output. Evaluating collaboration quality not just by what it produces today but by whether it makes the human more capable six months from now—a metric requiring longitudinal assessment.