Definition
Scaffolding is support that lowers what a task demands of you now while leaving you able to do it afterwards, or better able. Substitution is its opposite: the tool performs the operation and the need to do it yourself disappears. The same AI can do either. It scaffolds when it works like a tutor whose reasoning you rebuild and check. It substitutes when it hands you a finished synthesis you accept.
What is at stake
Both feel like help in the moment, and the output can look identical. The difference shows only when the tool is taken away, which is the one moment nobody tests. So the choice between them has to be built into how the work is set up, since it cannot be felt while working.
Where the word came from
- ScaffoldingWood, Bruner and Ross, 1976, the tutor who handles what the child cannot yet do, and whose help withers away as the child learns
- Cognitive artifactsDonald Norman, 1991, tools that support thinking
- Complementary against competitive toolsKrakauer, 2016, told apart by what is left when the tool is removed: the abacus leaves a mental model behind, the calculator leaves you where you started
- Scaffolding against substitutionSolé and colleagues, 2026, the same test applied to LLM use across a population
How episcope came to it
The word comes from a 1976 study of tutoring by David Wood, Jerome Bruner and Gail Ross, where the adult handles the parts of a task a child cannot yet manage and the help withers away as the child learns. The pairing with substitution, and the test of what remains when the tool is withdrawn, come from a 2026 model by Ricard Solé and colleagues of how LLM use spreads through a population. It builds on David Krakauer's distinction between complementary and competitive tools, and Krakauer is one of its authors.
What to watch for
Solé and colleagues' paper is a preprint that builds a mathematical model and reports no data, so it offers the distinction and the test rather than a measurement. A 2026 peer-reviewed study by Shaofeng Wang and Hao Zhang sets complementary cognitive processes against substitutive automation, but it measures learning by asking students, so it never takes the tool away to test what is left. The word scaffold also names one of the five roles in Scott's AI Cognitive Pyramid, where it describes what the AI does rather than what is left for the person. The two uses are easy to merge and better kept apart.
Papers and articles behind it
Wood, D., Bruner, J. S., & Ross, G. (1976). The role of tutoring in problem solving. Journal of Child Psychology and Psychiatry, 17(2), 89-100.
Krakauer, D. C. (2016). Will A.I. Harm Us? Better to Ask How We'll Reckon With Our Hybrid Nature. Nautilus, 5 September 2016.
Solé, R., Ruffini, G., Castaldo, F., Tuccio, M., Seoane, L. F., De Domenico, M., Elena, S. F., Krakauer, D. C., & Levin, M. (2026). Large-Language Models as a Cognitive Virus. arXiv:2609.03344.
Wang, S., & Zhang, H. (2026). Pedagogical partnerships with generative AI in higher education: how dual cognitive pathways paradoxically enable transformative learning. International Journal of Educational Technology in Higher Education.