How teachers can “vibe code” tools for students

In a year in which AI has continued to dominate the headlines, it’s perhaps no surprise that Collins Dictionary chose “vibe coding” as its word of the year for 2025[1].
If you haven’t come across this expression before, it’s defined as “AI-assisted software development”. To put it simply, it’s the process of building software by instructing an AI model in plain English rather than writing the code yourself.
What it has allowed is for anyone, even people with zero prior technical knowledge or experience, to build software. This has unleashed a new wave of creativity, with the App Store reporting 60% more apps released in December 2025 than the previous year[2].
For those of us working in the ELT industry, this represents a huge opportunity to build products to help students in a way that was never previously feasible.
For many ELT businesses, IT has traditionally been an afterthought — ensuring email accounts work and payroll runs smoothly rather than driving pedagogical innovation. Larger schools often found themselves navigating complex sales pitches from ed-tech providers, only to discover that implementation was far more complicated than promised.
Vibe coding for ELT
Vibe coding, when done correctly, allows any teacher or school to build custom software to address a very specific student problem.
For example, a teacher working with business English students could build a tool that generates role-play scenarios tailored to their students’ specific industry. A teacher preparing students for IELTS could create a writing feedback tool that analyses essays against the actual marking criteria, rather than relying on generic AI feedback. A school with a specific gap in their curriculum — say, a lack of good materials for C1 learners working on academic writing — could build something to fill it in a matter of hours rather than waiting for a commercial product that may never arrive.
This is no longer hypothetical. I never particularly enjoyed giving feedback or corrections on students’ written English, but I know how important high-quality feedback is for students. So I built FixMyEnglish.ai, a feedback and corrections tool designed specifically for non-native English speakers.
I had no prior coding experience. I described what I wanted to an AI model (Claude Code), refined the instructions when something didn’t work as expected, and iterated until I had a working product. The whole process felt less like programming and more like a conversation. Within a few days of evening tinkering, I had a tool that was live, functional, and ready to share with real learners. The entire thing took me less than five hours, and felt magical.
Opportunity for ELT
The barrier to entry has never been lower. For ELT professionals, who understand their students’ needs better than any Silicon Valley product team ever could, this is a huge opportunity. You don’t need to learn to code, hire a developer, or have any technical background whatsoever. What you do need is a clear idea of the problem you want to solve and a willingness to experiment.
Start by choosing an AI coding tool. Claude Code, Codex, and Replit are among the most popular options. My preference is for Claude Code, which requires a $20/month subscription, but most have free options which are more than sufficient for getting started.
The interface is remarkably similar to the one you’ll be familiar with in ChatGPT (and other AI chatbots). You simply type what you want, and the AI coding tool will build it for you. If you get stuck, just ask it to tell you what to do.
The key is in how you describe what you want to build. Forget the code, and focus on the problem and user experience. Who is going to use this tool? What will they type in or click on? What do you want them to get back? The more specific you are, the better the result. “Build me a vocabulary quiz” will produce something generic. “Build me a vocabulary quiz for B2 learners that presents a word in a sentence, offers four multiple choice answers, and gives an explanation of why the correct answer is right” will produce something genuinely useful.
When something doesn’t work — and there’s always something that doesn’t work — don’t panic. Describe the problem back to the AI in plain English, just as you would to a colleague in IT. The more information, the better.
This back-and-forth process of refinement is the core of vibe coding. It can be infuriating when the AI gets stuck, but these AI models are getting more capable by the day, and they are excellent at resolving complicated technical problems.
Finally, don’t aim for perfection on the first attempt. Build something small, test it with a small group of students, and improve it based on their feedback. The most useful tools often start as the simplest ones.
A few caveats are worth bearing in mind. Any tool that collects input from students, even just the sentences they type, raises data privacy questions. Before sharing anything with a class, check the terms of service of the platform you’re using and make sure that it complies with GDPR (or whichever regulations apply in your context). It’s also worth testing thoroughly before students get anywhere near it: AI makes mistakes, and a tool that produces subtly wrong grammar explanations or awkward example sentences can do more harm than good. Start small, with a trusted group, before rolling anything out more widely.
Good teaching has always been about understanding students’ needs, and tailoring the materials and approach to suit them. Vibe coding gives teachers, for the first time, the opportunity to build anything they want.
The only thing left is knowing what your students need. Fortunately, for now at least, that’s something no AI can tell you.
[1] https://www.bbc.com/news/articles/cpd2y053nleo
[2] https://the-decoder.com/vibe-coding-may-be-behind-a-60-percent-spike-in-new-ios-apps/





