#41. Educators learning how to work with AI can be a profound lesson; for us, our students and their parents
How I Built an AI-Powered Classroom Platform in Three Days (With No Coding or Technical Experience) and what it actually means.
Here is what I learned about building with AI — and what every educator and parent should understand about why learning to work with AI is now, possibly, as fundamental as learning to read.
1. AI Can Turn Educators Into Builders
The original version of this project lived in a Google Sheet with some App Script magic. It worked, but it was clunky. Students had to copy prompts, open ChatGPT in another tab, paste them in, then copy the AI responses back into their reports. The workflow was fragmented, and the experience felt more like admin work than an immersive simulation.
I wanted something better: a single platform where students could explore a mysterious island, interview AI citizens directly, submit reports, and collaborate as teams — all in one place. I did not know how to build that. I did not know what a "backend" was, what "tRPC" meant, or why "CORS errors" kept blocking the AI chat feature.
What I did know was pedagogy. I knew what P6 students in Hong Kong need: clear visual layouts, simple navigation, text-to-speech for accessibility, and English scaffolded at just the right level. I knew the narrative arc across eight sessions. I knew the four sectors, the twelve citizens, and the central mystery. The educational design was already fully formed in my head.
That is where AI comes in. Within hours, the platform had a landing page with an ocean-themed animated background, a "Read Aloud" feature that highlights text as it speaks, a fully rendered dashboard with sector cards and mission logs, and an integrated AI chat system where students type questions and citizens reply in real time. The AI built the entire codebase — React, TypeScript, database schema, API routes, the lot — while I focused on what students would see, read, and do.
The lesson: your subject expertise is the bottleneck, not the code. AI can handle the implementation. You bring the vision.
2. Strategic Thinking Beats Heavy Thinking
Here is the mistake people make: they think AI means you just describe a vague idea and get a finished product. Nothing could be further from the truth.
What actually happened over three days was a constant negotiation between what I wanted pedagogically and what the technology could reliably deliver. When the first preview loaded and the sector buttons did nothing, I had to diagnose why. When the iPad showed a blank screen, I had to figure out that CSS backdrop filters crash iOS Safari. When the AI chat returned CORS errors, I had to understand enough about cross-origin requests to ask the right questions.
None of that required writing code. But it required thinking strategically about where the problem lived — frontend, backend, API, or deployment — and giving the AI precise instructions about what to fix. The AI did the heavy lifting of rewriting database connections, migrating from MySQL to SQLite, restructuring API endpoints, and rebuilding the build pipeline. I did the light but strategic work of identifying which lever to pull.
The turning point came when we hit a wall with Render deployment. After multiple failed attempts at file creation through GitHub's web interface, I had to switch to Terminal commands, generate a GitHub Personal Access Token, and force-push the repository. That was not "pressing a button." That was thinking through authentication flows, understanding why GitHub Desktop got stuck in a fetch-publish loop, and persisting until the commit went through. The AI guided every command. I had to decide when to try a different approach.
The lesson: AI amplifies strategic thinkers. It does not replace them. The educators who will thrive are those who can decompose a problem, articulate the educational requirement clearly, and iterate based on what the technology reveals.
3. The Real Skill Is Learning How to Think With AI
I know there are parents and educators who look at AI with genuine concern. They worry that children will become dependent on it, that it will erode critical thinking, that it will replace creativity rather than enhance it. I understand those fears — AI can absolutely be used poorly. Students can paste essay prompts and copy the output without engaging their brains. That is not learning. That is outsourcing.
But here is what I discovered while building this platform: the most valuable skill I was developing was not technical. It was meta-cognitive. I was learning how to think about how I was using AI. I was learning when to trust the AI's first answer, when to push back, when to ask for a different approach entirely. I was learning to hold the educational goal in my mind while the AI proposed a dozen technical solutions — and to reject the ones that sounded impressive but would not serve my students.
That is the skill our students need. Not "how to use ChatGPT to write an essay." But how to collaborate with an AI system to achieve something that would be impossible alone. How to decompose a complex goal, assign the right tasks to the right tool, verify the output, and iterate. The students who grow up fluent in that workflow — who treat AI as a thinking partner, not a replacement for thinking — will be the most adaptable, productive, and employable adults in the workforce of the 2030s and beyond. (Teaching this skill may sound too grand an objective and inappropriate for younger learners. So I qualify this by pointing out that, of course, when we apply this thinking to what we ask our students to do, we must scaffold and make age and level appropriate any resources or tasks.)
We cannot simply say "AI is bad, let us ignore it." That is not a strategy. It is a surrender. The technology is not going away. The question is whether we teach our children to use it wisely, or leave them to figure it out unsupervised. Every hour I spent wrestling with CORS errors and GitHub authentication was an hour I spent modelling something for my students: persistence, strategic problem-solving, and the humility to ask for help when you are stuck.
4. Time and Patience Are Non-Negotiable
This project took three full working days. Not three hours. Three days of sustained focus, with moments of genuine frustration.
On day one, we got a basic prototype working — landing page, dashboard, static data. On day two, we hit the deployment wall: CORS errors, database crashes, missing environment variables, a 100-file GitHub upload limit, and a broken SQLite migration that took hours to trace. On day three, we rebuilt the entire deployment from a clean ZIP, wrestled with GitHub authentication, and finally got a successful force-push to Render. Then we added the Cards page with 36 investigation cards and embedded a YouTube introduction video.
The diagnostic checklist alone has 26 tests. Some will pass immediately. Others — particularly cross-device persistence — will fail because Render's free tier spins down after 15 minutes and the database connection is unstable. Those failures are not code bugs; they are platform limitations. Knowing the difference requires patience.
There were moments I wanted to quit. When GitHub Desktop refused to publish for the tenth time. When the API endpoint turned out to be
.cn instead of .ai. When the database schema push crashed because the disk mount was not ready. Each time, the fix was one precise edit — but finding that edit required stepping back, breathing, and working through the problem methodically.The lesson: AI accelerates development, but it does not compress it to zero. Budget real time. Expect dead ends. Trust that the next iteration will be closer than the last.
Bonus Takeaway: Build a Failsafe Into the Design
One of the smartest decisions we made was creating the Cards page — a completely offline, backend-independent gallery of 36 investigation cards. If the AI chat goes down, if the database connection drops, if Render's free tier is sleeping, students can still browse character portraits, sector information, clue cards, and event cards. The Cards page requires zero API calls, zero database queries, and zero authentication. It is a paper deck in digital form.
This is the kind of design thinking educators bring naturally. We know that technology fails in classrooms. We know that 30 students waiting on a loading screen is a lesson-killer. Building a graceful degradation path — where the platform still delivers value even when its most advanced features are unavailable — is a pedagogical instinct, not a technical one.
Final Verdict
The platform is live. Students can join a world, chat with AI citizens, submit reports, browse investigation cards, and watch an introduction video. The teacher can view all reports and manage the session. Is it perfect? No. Cross-device persistence is shaky on Render's free tier. The AI occasionally returns a fallback message on the first request. The database needs a paid tier to be truly reliable.
But is it usable in a classroom? Absolutely. And it was built by an educator who cannot code, working with AI as a collaborative partner.
If you are an educator with an idea, my advice is simple: start. Describe what you want. Test what you get. Break it, fix it, break it again. The AI will meet you halfway — but you have to bring the purpose, the persistence, and the pedagogical judgment that no algorithm can replicate.
Your students are waiting. Build it.
Comments
Post a Comment