Tool, tutor, or crutch?: A grounded theory of cognitive scaffolding and offloading in AI-assisted programming education
Open accessThe DOI matches its Crossref record; 36 of the 36 references it cites were found in a registry or an index; 29 of the 57 sources its claims cite were checked for support.
Abstract
Analysis revealed a core tension between students’ pursuit of “Domain Mastery” (conceptualization, explanation, and evaluation) and “Tool Mastery” (procedural efficiency with AI). We identified dynamic strategy switching (the Strategic Dance), Partnership Framing with an Illusion of Dialogue subtheme, and two recurrent evaluation challenges (Trust-but-Can't-Verify for novices; a Boilerplate Blindspot for more experienced students). We also describe attenuated meta-cognitive calibration–a miscalibration...
Overview
CoChat Summary
TL;DR
This constructivist grounded-theory study distinguishes Domain Mastery from Tool Mastery and proposes a tension model with Scaffolding and Offloading loops to explain how AI-assisted programming can improve performance while thinning germane processing and authorship. Across AI-enabled (N=24) and human pair-programming contrast (N=17) sections, it identifies Strategic Dance, Partnership Framing/Illusion of Dialogue, Trust-but-Can’t-Verify, Boilerplate Blindspot, and attenuated metacognitive calibration.
Key findings
- The core tension is between Domain Mastery (conceptualization, explanation, evaluation) and Tool Mastery (procedural efficiency with AI), with two recurrent loops—Scaffolding and Offloading—and boundary conditions including time pressure, task complexity/element interactivity, prior familiarity, and scaffolding design/norms.
- Students showed a Strategic Dance between Socratic Inquiry and Transactional Delegation: of 857 solution-seeking prompts, 75% (N=643) were Transactional Generation and 25% were Conceptual Inquiry; delegation typically increased within 24–36 h of deadlines, while novel APIs or multi-component tasks prompted renewed inquiry.
- Evaluation was uneven: novices showed Trust-but-Can’t-Verify, more experienced students showed a Boilerplate Blindspot, and Partnership Framing included an Illusion-of-Dialogue subtheme when partnership language co-occurred with one-shot acceptances; the human pair-programming contrast described social verification through rapid questioning and shoulder-surf review.
- Metacognitive outcomes included tool-linked self-efficacy and attenuated calibration: 14 of 24 AI users described pride in “prompt craft,” concept maps descriptively showed the human-pair contrast group with larger increases in cross-links and hierarchical organization, and some AI users expressed uncertainty about unassisted performance despite strong project grades.
Methods
- Constructivist grounded theory (Charmaz) with constant comparison, conducted over a 16-week Fall 2024 semester in a required second-year undergraduate Java course at a public university in Shanxi, China; two naturally occurring sections: AI-enabled GitHub Copilot in IntelliJ (N=24) and human pair-programming theoretical contrast (N=17), total 41 students.
- Data sources: IDE interaction logs (857 solution-seeking prompts; a stratified subset of 220 prompts coded for prompt type and verification intensity 0–2), pre-/post-course concept maps, and post-course 60-minute semi-structured dyadic interviews with stimulus recall; triangulation across data sources and analysts, supported by NVivo 14.
- Analysis proceeded through initial/open coding, focused coding with constant comparison, and theoretical integration via memoing and diagramming; member checking involved six participants (three AI-enabled, three pair-programming), and the study makes no causal or efficacy claims.
Limitations
- The paper states: sections differed in instructor/time/peer norms and were not constructed for controlled contrasts; the study covered a single 16-week semester at one Mandarin-language Chinese public university; concept maps are a narrow indicator; dyadic interviews can invite peer influence and social desirability; no individual follow-ups were conducted; interviews near exams confound anxiety with metacognitive accuracy; and the study claims theoretical sufficiency rather than universal saturation.
- The paper also states alternative explanations: cross-section contrasts may reflect social verification affordances of collaboration rather than AI per se; concept mapping offers only one window on conceptual structure; partnership language can be metaphorical; and pre-assessment anxiety may partly explain reported uncertainty.
- Evident from the text: the small qualitative sample and non-equivalent ecologies limit generalizability, the concept-map comparisons are descriptive rather than inferential, and the design cannot support causal claims.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
19 Valid · 1 Not valid · 17 Not checked · 153 Own results and interpretations
Some cited sources support this claim and some don’t — read each source’s verdict below.
In the paper
“inquiry phases tend to reduce extraneous effort while retaining germane processing”
What the sources say
- Partly supportedcites Debue & Van De Leemput, 2014 sourceExperimental — accuracy not yet measured
The Introduction’s Cognitive load theory section says extraneous load should be kept low to leave mental resources available for learning, and the Research questions section expects higher extraneous load to reduce resources for germane processing. But the paper studies versions of an online newspaper, not inquiry phases; it does not show that such phases tend to produce this effect. The indexed copy provides section labels but no page numbers. (LLM, conf=0.82)
- Disputedcites Stadler et al., 2024 sourceExperimental — accuracy not yet measured
The Discussion (p. 5) reports that students using LLMs had lower extraneous and germane cognitive load than students using traditional web searches. Thus, although the paper supports reduced extraneous effort with LLM use, it says germane processing was also reduced, not retained. The study compares search methods rather than inquiry phases. (LLM, conf=0.87)
Checked by a reasoning model — may be wrong.
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“Empirical evidence suggests that AI assistance can improve short-term performance.”
What the sources say
- Not checked — no full text of the sourcecites Yilmaz & Yilmaz, 2023 source
- Supportedcites Lyu et al., 2024 source
Section 4.1 (p. 6) reports significantly higher scores on components where CodeTutor was allowed than on components where it was not. This supports the hedged claim that AI assistance can improve short-term performance, in the context of introductory programming coursework. (LLM, conf=0.88)
- Supportedcites Fan et al., 2025 source
The Results report that AI-assisted groups outperformed both individual and human–human groups on programming tasks. The Discussion describes immediate gains in task completion speed and initial performance, supporting the claim's short-term qualifier. (LLM, conf=0.94)
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“AI tools often reduce extraneous load (less boilerplate search, fewer syntax stalls) (Jose et al., 2025).”
What the sources say
- Partly supportedcites Jose et al., 2025 sourceExperimental — accuracy not yet measured
In “Mapping AI applications to theoretical constructs” (page unavailable), the paper says AI can eliminate redundant work and let students focus on more important cognitive operations. It describes this as a potential benefit, not something that often occurs, and does not mention boilerplate searches or syntax stalls. The passage also says “maximize the extraneous load,” so it does not clearly state the claimed reduction. (LLM, conf=0.78)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Studies now warn that students who rely heavily on GenAI often struggle to articulate why autogenerated code works (Jose et al., 2025) and score lower on delayed conceptual tests (Bastani et al., 2024).”
What the sources say
- Partly supportedcites Jose et al., 2025 sourceExperimental — accuracy not yet measured
The 'Addressing AI pitfalls in education' section (page unavailable) warns that students who over-rely on AI answers might bypass critical thinking and self-reasoning, and recommends asking them to describe AI-given answers in their own words. It does not discuss autogenerated code or report that students often struggle to explain why such code works. (LLM, conf=0.85)
boostedstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“First, for high element interactivity tasks, direct solutions can obscure necessary coordination work, inviting premature acceptance and thus limiting germane processing (Stadler et al., 2024).”
What the sources say
- Partly supportedcites Stadler et al., 2024 sourceExperimental — accuracy not yet measured
The Discussion (p. 5) reports that students using LLMs received direct, concise answers and experienced lower germane cognitive load than students using web searches. Section 2 (p. 2) describes high element interactivity as requiring integration of multiple components. The paper does not establish that direct answers obscured coordination work or caused premature acceptance; its Limitations (p. 5) note that students’ cognitive processes were not directly observed. (LLM, conf=0.82)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
61 more claims · 8 with their cited sources checked
Create a free account to see all 190 claims and to verify the 1 reference still open.
- Probably not a substantial claim. 74
- It may state two separate claims as one. 78
- The quote only partly states this claim. 1
Citations
Cites
The Discovery of Grounded Theory: Strategies for Qualitative Research
Glaser, B. G., & Strauss, A. L. (2017) · Aldine (Routledge reprint) · doi:10.4324/9780203793206 · Cited by 25,789
Interviews – Learning the Craft of Qualitative Research Interviewing
Vaivio, J. (2012) · European Accounting Review · doi:10.1080/09638180.2012.675165 · Cited by 4,535
Niemiec, C. P., & Ryan, R. M. (2009) · Theory and Research in Education · doi:10.1177/1477878509104318 · Cited by 2,542
Chiu, T. K. F., Xia, Q., Zhou, X., Chai, C. S., & Cheng, M. (2023) · Computers and Education: Artificial Intelligence · doi:10.1016/j.caeai.2022.100118 · Cited by 1,181
Expertise Reversal Effect and Its Implications for Learner-Tailored Instruction
Kalyuga, S. (2007) · Educational Psychology Review · doi:10.1007/s10648-007-9054-3 · Cited by 1,056
Kapur, M. (2008) · Cognition and Instruction · doi:10.1080/07370000802212669 · Cited by 867
Yilmaz, R., & Karaoglan Yilmaz, F. G. (2023) · Computers and Education: Artificial Intelligence · doi:10.1016/j.caeai.2023.100147 · Cited by 745
Teacher support and student motivation to learn with Artificial Intelligence (AI) based chatbot
Chiu, T. K. F., Moorhouse, B. L., Chai, C. S., & Ismailov, M. (2024) · Interactive Learning Environments · doi:10.1080/10494820.2023.2172044 · Cited by 707
Xia, Q., Chiu, T. K. F., Lee, M., Sanusi, I. T., Dai, Y., & Chai, C. S. (2022) · Computers & Education · doi:10.1016/j.compedu.2022.104582 · Cited by 469
Shabani, K., Khatib, M., & Ebadi, S. (2010) · English Language Teaching · doi:10.5539/elt.v3n4p237 · Cited by 458
Research works printed but not linked: 6
- Charmaz, K. (2014). Constructing grounded theory (2nd Ed). SAGE.
- Collins, A., Brown, J. S., Holum, A., et al. (1991). Cognitive apprenticeship: Making thinking visible. American Educator,15(3), 6–11.
- Leinwand, S., Brahier, D. J., Huinker, D., Berry, R. Q., Dillon, F. L., Larson, M. R., & Smith, M. S. (2014). Principles to actions: Ensuring mathematical success for all. National Council of Teachers of Mathematics: NCTM.
- Lincoln, Y. S., & Guba, E. G. (1988). Criteria for assessing naturalistic inquiries as reports. New Orleans, LA: Annual meeting of the american educational research association.
- Patton, M. Q. (2002). Qualitative research and evaluation methods (vol 3). Sage.
- Vygotsky, L. S. (1978). Mind in society: The development of higher psychological processes (vol 86). Harvard University Press.
Cited by
Qian, W., Yang, F., Cao, Y., Yi, L., Gu, R., & Wang, Z. (2026) · Frontiers in Psychology · doi:10.3389/fpsyg.2026.1921504 · Cited by 4
“The clearest case is the grounded-theory study by Liu D. et al. (2026), whose title— tool, tutor, or crutch—names the spectrum directly and whose analysis states that ‘offloading is evident when both extraneous and germane processing are minimized,’ precisely the configuration in which a falling load…”
Liu, D., Fan, G. F., & Pan, L. (2026) · International Journal of STEM Education · doi:10.1186/s40594-026-00611-4
Self-citationSchadow, F., Gerling, K., Kretzer, F., & Maedche, A. (2026) · Annual Symposium on Human-Computer Interaction for Work (CHIWORK) · doi:10.1145/3805029.3818286
Information Literacy Through Open Pedagogy: What’s the Role of AI Tools?
Clinton‐Lisell, V. (2026) · Sustainability · doi:10.3390/su18178684
Mokha, A. K. (2026) · Development in Learning Organizations An International Journal · doi:10.1108/dlo-06-2026-0321
Metadata
- Authors
- Liu, D., Fan, G., & Pan, L.
- Affiliations
- University of Malaya (MY)
- Taiyuan University of Science and Technology (CN)
- Corresponding author
- Liu, Dandan
- Fan, Guangrui
- Pan, Lihu
- Venue
- International Journal of STEM Education
- Published
- 2026-03-13
- Volume
- 13
- Issue
- 1
- ISSN
- 2196-7822 (ISSN-L)
- OpenAlex
- W7135212788
- Cited by
- 16
- References
- 27
- Citations by year
- 2026: 16
- FWCI
- 133.49
- Type
- journal article
- Language
- en
- Update notices
- correction (2026-04-07)
- Open access
- gold
- doi.org
- Licence
- cc-by-nc-nd
- creativecommons.org/licenses/by-nc-nd/4.0 (tdm)
- creativecommons.org/licenses/by-nc-nd/4.0 (vor, 3 d)
- Locations
- International Journal of STEM Education (publishedVersion, cc-by-nc-nd)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- Natural Science Foundation of Shanxi Province (No.202203021221145)
- Shanxi Province Intelligent Software and Human Machine Environment System Postgraduate Joint Training Demonstration Base (No.2022JD11)
- Topics
- Physical Sciences › Computer Science › Computer Science Applications › Teaching and Learning Programming
- Physical Sciences › Computer Science › Artificial Intelligence › Intelligent Tutoring Systems and Adaptive Learning
- Social Sciences › Education › Artificial Intelligence in Education
- Keywords
- cognitive scaffolding, offloading, AI-assisted programming, conceptual understanding, transfer of learning, pair programming, trust in AI, boilerplate code, self-determination theory, cognitive load theory, AI authorship, Java programming
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Tool, tutor, or crutch?: A grounded theory of cognitive scaffolding and offloading in AI-assisted programming education. https://app.cochat.ai/papers/10.1186/s40594-025-00592-w
- The paper
- Liu, D., Fan, G., & Pan, L. (2026). Tool, tutor, or crutch?: A grounded theory of cognitive scaffolding and offloading in AI-assisted programming education. International Journal of STEM Education. https://doi.org/10.1186/s40594-025-00592-w
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.