Vibe coding made the Collins Word of the Year list for 2025, and suddenly everyone had an opinion. Too risky. Too sloppy. Not real development. The critics lined up quickly.
But here is the thing they keep missing: most of the people doing vibe coding in education are not trying to ship software. They are trying to help a Year 8 student understand how a CPU fetch cycle works before period three ends.
That changes the calculation entirely.
What vibe coding actually means
Collins defines vibe coding as using artificial intelligence, prompted through natural language, to help write code. Andrej Karpathy coined the term earlier in 2025, and it stuck because people recognised themselves in it. We were already talking to our tools. Already describing outcomes instead of grinding through every bracket and semicolon.
The debate that followed split into two camps. One side argued for "flow coding," where you build structured workflows to make AI-generated code stable enough for production. The other side stayed on the vibe, building small things quickly and moving on.
I sit firmly on the vibe side. Not because I think flow coding is wrong. It makes perfect sense if you are building larger systems. But I am not building larger systems. I am teaching.
Why vibe coding fits how teachers actually think
Teaching is full of moments where you can feel the frustration in the room. Students trying to imagine the abstract. What a CPU fetch cycle looks like even though they cannot see inside a processor. How packets travel across the internet. How a wireless signal finds its way across a crowded classroom.
This sits across the entire curriculum. Tectonic plates, the water cycle, blood flow, predator and prey. Key Stage 3 is full of beautifully complicated processes that students are expected to understand, even though they cannot see them.
Vibe coding gives me the chance to say: let me build something that makes the idea visible.
Not a full application. Not a polished tool. Just a small simulation or interactive that lets students see what is usually hidden. And because it is quick to produce, the lesson can shift in real time.
A few examples from my own practice:
- A packet-travelling model where each hop becomes visible
- A CPU cycle visualiser showing fetch, decode, execute step by step
- A particle simulation to make heat transfer feel physical
- A tectonic plate model that moves slowly and predictably
- A simple ecosystem showing population changes over time
But it does not stop at simulations.
Vibe coding lets me build small tools that help students practise more independently. Interactive questions with revealable model answers. Revision drills. Mastery dashboards in Canva where students click on a question number and instantly see targeted feedback from a bank I have written myself.
None of this is glamorous. None of this is commercial. It makes a difference in my classroom, and that is enough.
Where the data boundaries sit (and why they matter)
I do not collect data from any vibe-coded creation unless it lives entirely inside Canva and connects to a Canva Sheet. That is the one safe exception because everything remains inside the same school-friendly environment.
Anything built in Google AI Studio is a one-shot, standalone app. No login. No storage. No data held anywhere.
One important note: AI Studio trains on your prompts when coding, so you have to be careful. No personal data. No student details. No copyrighted material that should not be shared. The model learns from what you type, so think before you type.
Keeping these boundaries is not about being cautious to the point of paralysis. It is about respecting the legal and ethical landscape. It keeps me out of the GDPR quagmire that I have no desire to enter.
Knowing your limits is what makes it workable
People worry about buggy code, insecure outputs, and hidden vulnerabilities in AI-generated projects. I understand that. The concerns are legitimate when you are building software that other people depend on.
Which is why I stay firmly in the territory where those risks barely register. My work is small, contained, and short-lived. I am a teacher who AI-codes, not a developer preparing a product for the market.
I know my limits. The limits of the vibe. The limits of the flow.
Staying inside those limits means vibe coding becomes a supportive tool instead of a risk.
How to start if you want to try this yourself
If you are curious, start small. Here is a practical starting point:
- Use Canva code for simple interactives and feedback tools. It is familiar, school-friendly, and keeps everything contained.
- Use Canva Sheets only when you need minimal data inside a secure space. No external databases, no third-party integrations.
- Use AI Studio for standalone visualisations that store nothing. One-shot apps with no data persistence.
- Pick one difficult concept and model it in a small, clear way. A single simulation that makes the invisible visible.
Keeping your first steps modest is not a compromise. It is the safest and most productive way to learn what this approach can offer.
The real point: clarity, not products
Vibe coding, at least for me, is not about creating products. It is about creating clarity.
It is a way to bring abstract ideas within reach and build small moments of understanding. Consider the possibilities across subjects:
- A fractions visualiser where students drag fraction bars to create equivalents
- A photosynthesis toggler where students adjust light levels and see plant growth indicators change
- A heart circulation diagram with tappable arrows showing blood movement
- An angles explorer where students rotate lines and watch measurements update
- A carbon cycle simulator where students move carbon between atmosphere, plants, animals, and oceans
None of these are publishable products. They are tiny, purposeful tools, and the only audience that ever sees them is the class in front of me.
Vibe coding works when you stay on the vibe side. When you keep things small. When you refuse to collect data you do not want to handle. When you use just enough technology to help students finally see the thing they have been trying so hard to imagine.
Matthew Wemyss is an AIGP-certified AI in Education consultant and practising school leader. Book a discovery call to discuss building custom AI tools for your school.
