Vibe-Lesson-Planning: When AI Becomes Your Co-Teacher Before Class Even Starts
Borrowing from "vibe coding", where developers delegate code generation to AI with minimal review, this article introduces vibe-lesson-planning: the emerging practice of co-creating lessons with generative AI. Drawing on recent scholarship, it offers educators three practices to follow, briefing AI with rich contextual detail, iterating rather than accepting first drafts, and using AI as critic, alongside three risks: context-blind content, cognitive offloading of pedagogical thinking, and homogenised lessons stripped of local texture. Authorship, it argues, must remain with the teacher.
In February 2025, the computer scientist Andrej Karpathy posted a short note on X that gave a name to something programmers were already doing. He called it vibe coding: describing what you want in plain language, letting an AI write the code, and barely reading the output (Karpathy 2025). The term spread so fast that researchers are now studying it formally, defining it as a fundamental reconfiguration of how human intent gets translated into working systems, with the AI acting as a pair programmer rather than a mere tool (Meske et al. 2025).
I want to borrow the term, because teachers are already doing the same thing. They are just not calling it anything yet.
Call it vibe-lesson-planning: describing tomorrow's lesson to an AI in plain language, letting it draft the objectives, activities, and worksheets, and walking into class with the output. The research confirms what staffrooms already know. Generative AI can produce usable lesson plans and support materials in minutes, and studies in teacher education have found genuine potential in these tools for planning, creativity, and even critical thinking, provided they are used deliberately (van den Berg & du Plessis 2023). The operative word is deliberately. Because just as vibe coding produces software that runs but that nobody fully understands, vibe-lesson-planning can produce lessons that look complete but that nobody has actually thought through.
Having spent the past few years training Malaysian educators in AI, and watching what separates the teachers who thrive with these tools from those who quietly get worse, I offer three practices to follow and three traps to watch.
Follow this: brief the AI like you would brief a colleague.
The single biggest difference between a generic lesson plan and a usable one is context. Research on how teachers actually prompt these models shows that outcomes depend heavily on the specificity and quality of what teachers put in, not just what the AI puts out (Moorhouse et al. 2025). Do not ask for "a lesson on fractions". Instead, try:
"You are my co-planner. Design a 60-minute Form 2 Mathematics lesson on equivalent fractions for a mixed-proficiency class of 34 students in a Malaysian daily school. I have one projector, no student devices, and the last period on a Thursday. Align it to the KSSM curriculum. Include one low-floor activity for struggling students."
Every constraint you state is a bad suggestion you prevent.
Follow this: iterate, never accept.
Vibe coders click "accept all" on code they have not read. Do not be that teacher. Treat the first output as a draft from an eager but inexperienced trainee. Push back:
"Activity 2 assumes my students already understand denominators. They do not. Redesign it with more scaffolding, and replace the worksheet with a think-pair-share task that gets them talking in Malay or English, whichever they are comfortable with."
The magic of these tools is not in the first response. It is in the third and fourth, after you have argued with it.
Follow this: make the AI your critic, not just your clerk.
The most underused prompt in education is the one that asks AI to challenge you rather than serve you. Once your plan is ready, flip the roles:
"Act as a critical colleague reviewing this lesson plan. Question whether my activities actually deliver my stated learning objectives. Identify where weaker students will get lost, and where my assessment fails to measure what I claim to be teaching."
This turns planning back into thinking, which is where it belongs.
Now, the three traps.
Watch out: the AI does not know your students, and it will not tell you that.
These systems produce confident, polished plans regardless of whether the content is accurate or appropriate. Reviews of AI applications for teachers consistently note that these tools fall short of understanding real classroom needs and contexts (Celik et al. 2022). The AI has never met the boy at the back who reads two grades below level, and it does not know that your school's science lab has been under renovation since March. It will also occasionally invent facts, misattribute sources, and misread the local curriculum with total confidence. Verify every factual claim and every curriculum alignment yourself. The plan is a draft. You are the fact-checker of record.
Watch out: the planning was never the paperwork. It was the thinking.
Here is the uncomfortable part. Emerging evidence links heavy reliance on AI tools with weakened critical thinking through cognitive offloading, the gradual outsourcing of mental effort to the machine, with younger users most affected (Gerlich 2025). Lesson planning has always been a thinking discipline disguised as a writing task. It is where you mentally rehearse the lesson, anticipate misconceptions, and sequence ideas. If the AI does all of that while you merely skim, you arrive in class holding a plan your own mind has never visited. The vibe coder ships code they cannot debug. The vibe planner teaches a lesson they cannot adapt when it goes sideways at 8.15 in the morning. And lessons always go sideways.
Watch out: the quiet homogenisation of your classroom.
Left unbriefed, these models default to a generic, vaguely Western classroom that exists nowhere. Ask ten teachers for an AI lesson on photosynthesis and you will receive ten nearly identical plans, none of which mention a pokok bunga raya or connect to what students saw on the way to school. Over time, unedited AI plans sand away the local texture that makes lessons land. Your context is your professional advantage; the AI only has it if you insist on it, every single time. And while we are at it: never paste student names, marks, or personal details into these tools. Convenience is not consent.
So is vibe-lesson-planning good or bad? That is the wrong question, the same way it was the wrong question for calculators and for Google. The better question is who remains the author. Used well, the AI is a tireless co-planner that handles the formatting, drafting, and heavy lifting while you supply the judgement, the context, and the intent. Used lazily, it is a photocopier of other people's pedagogy, and your professional thinking quietly atrophies with every accepted draft.
Karpathy, to his credit, said vibe coding was meant for throwaway weekend prototypes, not production systems. Our classrooms are production systems. Our students are not prototypes.
Vibe all you want at the drafting stage. But when you walk into that classroom, make sure the lesson is yours.
References
- Celik, I., Dindar, M., Muukkonen, H., & Järvelä, S. (2022). The Promises and Challenges of Artificial Intelligence for Teachers: a Systematic Review of Research. TechTrends, 66(4), 616–630. https://doi.org/10.1007/s11528-022-00715-y10.1007/s11528-022-00715-y
- Gerlich, M. (2025). AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(1), 6. https://doi.org/10.3390/soc1501000610.3390/soc15010006
- Karpathy, A. (2025). There’s a new kind of coding I call “vibe coding.” X. https://x.com/karpathy/status/1886192184808149383
- Meske, C., Hermanns, T., Von der Weiden, E., Loser, K.-U., & Berger, T. (2025). Vibe Coding as a Reconfiguration of Intent Mediation in Software Development: Definition, Implications, and Research Agenda. IEEE Access, 13, 213242–213259. https://doi.org/10.1109/access.2025.364546610.1109/access.2025.3645466
- Moorhouse, B. L., Ho, T. Y., Wu, C., & Wan, Y. (2025). Pre-service Language Teachers’ Task-specific Large Language Model Prompting Practices. RELC Journal. https://doi.org/10.1177/0033688225131370110.1177/00336882251313701
- van den Berg, G., & du Plessis, E. (2023). ChatGPT and Generative AI: Possibilities for Its Contribution to Lesson Planning, Critical Thinking and Openness in Teacher Education. Education Sciences, 13(10), 998. https://doi.org/10.3390/educsci1310099810.3390/educsci13100998