Is hands-on learning still necessary in the age of AI? A thematic review

Synthesis: A thematic review of 32 studies argues that generative AI does not eliminate hands-on learning in design education but reorganizes its forms into two complementary strands: Embodied Hands-on (bodily action, tools, materials) and Cognitive Hands-on (language, judgment, and human–AI iteration). Both share a cycle of action, Feedback, reflection, and refinement grounded in Experiential Learning and Embodied Learning, yet rely on different sources of feedback and are not educationally interchangeable, with rapid generation raising concerns about compressed Creativity processes and growing Human AI Collaboration demands on evaluative judgment and Metacognition.

Key Findings

• AI reorganizes rather than replaces hands-on learning. The reviewed literature shows sketching, model making, material experimentation, and prototyping remain core activities, while AI-supported generation, comparison, and selection become a second, cognitive form of practice.
• Two distinct forms emerge from the review. Embodied Hands-on relies on bodily action, tools, and materials; Cognitive Hands-on develops through continued operation, judgment, and adjustment of AI-generated outcomes — a distinction that emerged from the coding rather than being predetermined.
• Both forms share one process structure. Action, feedback, reflection, and refinement are the common cycle, but the two depend on different sources of feedback (real-world material resistance vs. language and visual outcomes) and should not be treated as equivalent.
• Rapid generation may compress learning processes. Fleischmann (2025, 2026) and others report that generation efficiency can reduce exploratory sketching and gradual trial-and-error, lowering opportunities for students to encounter failure and material constraints; more iterations enabled by AI do not necessarily mean greater iterative depth.
• AI use alone does not directly improve creativity. Zhu et al. (2025a) found that prompting without further steps did not raise creativity, whereas multi-step operations (generation, modification, selection) were positively associated with the creativity of final work.
• Cognitive Hands-on elevates metacognition and language. Students must translate design intentions into prompts, judge alignment between outputs and goals, and reflect on their own judgment processes — but language cannot replace material feedback, and over-evaluation may narrow tolerance for uncertainty and risk-taking.
• Design literacy is being redefined. The review proposes that design literacy increasingly involves critical evaluation, metacognitive regulation, and decision-making within human–AI collaboration, centered on learner agency.

Study Design & Method

The study used an inductive thematic review approach to examine how the role of hands-on learning is changing in AI-supported design education. Literature was identified through Web of Science and Scopus, with two search pathways: practice-based/studio learning traditions and AI-supported design learning. Screening narrowed an initial pool of records to a final corpus of 32 peer-reviewed studies. Analysis followed Braun and Clarke's (2006) thematic analysis in three stages — open coding, focused coding, and theme development — conducted in ATLAS.ti, with first-author-led coding and partial review by the third author; inter-coder reliability was not computed. Study quality was appraised using MMAT principles, and the PRISMA 2020 framework supported transparent reporting of identification and screening.

What this means for practice

• Instructors. Require the full generate, modify, select sequence on AI-supported design tasks instead of accepting a first generated output: the review reports that prompting alone did not raise the creativity of final work while multi-step operations did. Pair that prompt instruction with an explicit account of what design experience cannot be verbalized, so students do not drift into over-reliance on easily describable proposals.
• Instructors. Keep physical making in every project and route generated proposals back to it for testing, because material resistance supplies feedback that visually plausible outputs do not, and the review's central claim is that the two forms are not educationally interchangeable.
• Designers. Sequence embodied and cognitive hands-on in curriculum design against learning objectives rather than defaulting to efficiency, deciding in advance which AI-generated outcomes must be built and tested physically and which can stay in the language-and-judgment loop.
• Designers. Assess the action, feedback, reflection, and refinement cycle itself rather than counting generated outcomes or final-work scores, since learning gains depend on how students interpret, judge, and use feedback — the point the review makes for both teaching and assessment.
• Researchers. Follow students past a single course or short project: the review found most evidence stops at within-course performance, so the long-term effects of increasing cognitive hands-on remain unknown.

Limitations

• Conceptual, not yet empirically tested. The Embodied/Cognitive Hands-on framework has not been empirically validated, and there is insufficient evidence on whether the two forms can replace one another.
• Inconsistent definitions and measurement. Included studies use different conceptualizations of hands-on learning and divergent indicators (final-work scores, number of generated outcomes, interaction records) that may not reflect process quality.
• Limited longitudinal evidence. Most studies capture performance within a single course or short project, so long-term effects of increasing Cognitive Hands-on remain unclear.
• Narrow database reliance. The review primarily used Web of Science and Scopus (ERIC only in scoping), which may have limited comprehensiveness and excluded relevant studies indexed elsewhere.

Connected Concepts

• Experiential Learning — hands-on practice is grounded in Kolb's cycle of experience, reflection, conceptualization, and action.
• Embodied Learning — embodied cognition theory explains why material feedback and bodily action support design judgment.
• Creativity — the paper examines how AI generation, comparison, and revision reshape creative thinking via the creative-cognition framework.
• Human AI Collaboration — prompting, generation, selection, and revision constitute the cognitive strand of hands-on learning.
• Metacognition — students must monitor their own judgment of AI outputs, not merely the outputs themselves.
• Design Thinking — the study is situated in studio-based design pedagogy and designerly ways of knowing.
• Active Learning — hands-on participation via action, feedback, and iteration reflects constructivist active engagement.
• Pedagogies and Teaching Strategies — findings are framed as pedagogical tensions and curriculum design guidance for design education.

Connected Articles

• Artificial Intelligence as Catalyst and Contested Terrain: Transforming Interior Design Practice, Pedagogy, and Professional Regulation in Malaysia — examines generative AI integration in a design discipline, extending the same design-education context.
• Generative AI and Extended Reality in Collaborative Architectural Design Education: An Exploratory Studio Study — addresses AI in architectural design education, a primary field within the review corpus.
• Enacting Constructive Conflicts with AI Agents to Enhance Reconsideration among Novice Interaction Designers — explores learner–AI interaction and agency in design education.
• AI Personas: Can LLMs Replace Fieldwork in Experiential Learning? — investigates AI within experiential, practice-based learning settings.
• A study of GenAI usage by Design Students: Analysis of Survey Results and Journals of AI practices at the Politecnico di Milano in 2025/2026 — surveys design students' actual generative AI use, complementing the review's synthesis.

Citation

Yu, J., Liu, L., & Zhu, Q. (2026). Is hands-on learning still necessary in the age of AI? A thematic review. Frontiers in Psychology, 17, 1897168.