With a Thermomix You Lose the Ability to Cook: A Kitchen Machine Analogy for Applications of Generative AI in Education

Synthesis: Rummel, Nachtigall, and Panadero use the analogy of the Thermomix — a smart kitchen appliance whose guided cooking mode provokes the same enthusiasm and critique as generative AI in education — to show that the central question is not whether learners use AI but how that use shapes their learning. Mapping four cooking scenarios onto learning-with-AI cases within the ICAP and SAMR frameworks, they illustrate a progression from passive substitution (full delegation) to interactive redefinition (AI as a dialogue partner for engagement and co-construction). The paper argues against polarized media-comparison designs and urges researchers to disentangle the specific conditions under which generative AI fosters — or hinders — productive learning.

Introduction

The paper opens by noting that, just as cooking lies on a continuum between preparing a meal entirely from scratch and ordering fast food, learning with AI spans a spectrum between doing all the work unaided and relying entirely on the tool. The Thermomix provides a culturally resonant, emotionally charged analogy that moves the debate away from "AI good or bad" toward how tool use shapes learning processes, identity, and agency.

The ICAP and SAMR frameworks

Two established frameworks organize the paper's four scenarios. The ICAP framework (Chi & Wylie, 2014) distinguishes four modes of cognitive engagement — Passive, Active, Constructive, and Interactive — with higher modes assumed to produce superior learning outcomes. The SAMR model (Puentedura, 2006) distinguishes Substitution, Augmentation, Modification, and Redefinition — the extent to which technology enhances and transforms learning.

Four scenarios of use

The paper pairs four Thermomix uses with corresponding learning-with-generative-AI cases:

1. "With a Thermomix you lose the ability to cook" — Guided cooking and fully outsourcing assignments without revision or critical engagement. Aligns with ICAP Passive and SAMR Substitution; risks skill loss, limited creativity, and over-reliance (e.g. novice writers who revise AI-produced text only superficially).
2. "As a cook I can alter or add ingredients to some extent" — Users with prior knowledge adapt recipes; students refine prompts, cross-verify AI outputs with sources, and revise text based on prior knowledge. Aligns with ICAP Active and SAMR Augmentation; requires prerequisite knowledge and self-regulation.
3. "I have become a more creative and versatile cook" — Experienced users employ manual functions; students use AI to brainstorm, outline, evaluate, and build original work, fostering Creativity and reflection. Aligns with ICAP Constructive and SAMR Modification; positions AI as a "sparring partner" for meaning-making.
4. "A new way of cooking" — Online Thermomix communities and co-construction; students interact with AI as a dialogue partner providing real-time feedback and adaptive feedback. Aligns with ICAP Interactive and SAMR Redefinition; enables personalized, adaptive learning at scale.

Identity, agency, and evaluation

A key contribution is the reflection on what makes someone a cook — or a learner. Does authentic competence require carrying all knowledge internally, or can it include mobilizing external resources including generative AI? Drawing on Cox (2024), the paper suggests AI may shift learners from makers of their own knowledge to managers of externally provided knowledge, requiring complex cognitive skills to understand, evaluate, and organize information. It also raises Learner Agency and Assessment questions: educators, like diners judging a meal, often evaluate the final product without insight into how much AI contributed — prompting calls for assessment methods integrated into learning with AI.

Conclusion

The authors argue the key is not whether students use AI but how that use shapes the learners they become. They critique media-comparison designs for confounding factors and limited insight, and urge future research toward a deeper understanding of the specific conditions under which generative AI fosters productive learning processes — moving beyond polarized narratives toward nuance.

What this means for practice

• Educators. Classify the task by engagement mode before counting it as an improvement: a student who revises AI-generated text superficially is in the Passive/Substitution cell however fluent the output looks.
• Educators. Place cross-verification and prompt-refinement tasks where students have the prerequisite knowledge and self-regulation they demand — the paper ties the "alter the recipe" scenario to knowing AI's limits, judging credible sources and comparing across references.
• Instructional designers. Aim tasks at the Constructive and Interactive end — brainstorming, outlining, evaluating and building original work with AI as a sparring partner or dialogue partner — rather than guided delegation of the whole product.
• Instructional designers. Design assessment that makes the process visible, since educators see the final product without knowing how much AI contributed, which is where identity and agency questions get decided.
• Researchers. Stop running media-comparison designs that confound factors and instead isolate the conditions under which generative AI supports productive learning, and for whom.

Limitations

• The paper is conceptual: a kitchen-appliance metaphor plus four illustrative scenarios, with no participants, intervention or outcome data.
• The mapping of Thermomix uses onto the ICAP and SAMR cells is the authors' own construction, so the four scenarios are illustrations rather than observed or tested learning cases.
• What the scenarios are said to require — prior knowledge, self-regulation, the ability to judge credible sources — is drawn from cited prior work rather than measured here.
• The authors present the domains as analogous rather than equivalent, describing the appliance as "sufficiently distant" to permit reflection, so the transfer of the analogy to learning remains an argument rather than a finding.

Citation

Rummel, N., Nachtigall, V., & Panadero, E. (2026). With a Thermomix You Lose the Ability to Cook: A Kitchen Machine Analogy for Applications of Generative AI in Education. arXiv preprint arXiv:2609.09856.