Building my son a Minecraft phonics game with AI
My son is learning to read and write, and has a phonics test coming up. Like most kids, phonics practice isn't always his favourite thing — it's hard work, and hard work requires motivation. He also loves Minecraft with a level of dedication that I can only describe as professional.
So I built him a Minecraft phonics game in just one conversation with Claude AI.
The build
I opened a conversation with Claude and described what I wanted: a simple browser game to help a child practice phonics, Minecraft-themed, with two modes — one where he sees a picture and has to spell the word using letter tiles, one where he reads a word and picks the matching picture from four options. I described that the words should be appropriate for a child at Key Stage 1 in the UK curriculum.
What came back in a single response was a working React component styled to look like Minecraft with two playable game modes:
The iteration
Because I'm an engineer and it's hard to leave things alone, I took it further. I pulled the file into a proper project using Claude Code — set up Vite, added Vitest, switched to TypeScript. I found an npm package with real Minecraft item assets and swapped out the emoji placeholders for the actual in-game sprites. I added difficulty levels so the word pool scales as he improves.
TypeScript makes the data structures explicit, and with tests I can add features without worrying I've broken something. The real assets helps make the icons much more recognisable, some of the emoji choices were a bit questionable.
While this gave some improvements, it wasn't required to get value from it. The first version was playable, Minecraft-themed, and covered the right phonics content. A parent with no engineering background could have just used the playable version within Claude's UI. Even the changes I did make were just a few prompts with Claude Code.
The reaction
When I explained the game to my son he was instantly excited, his first reaction getting out of school was to eagerly ask if he could play it. He's been regularly asking to play it in the days since.
What I didn't anticipate was the co-design. He started giving feedback almost immediately - that we really should include netherite since it makes the strongest armour and would be a tricky word. He was invested in it and the phonics practice was happening alongside something he actually wanted to do and felt ownership of.
There were side effects I hadn't considered either. He hasn't had much experience using a computer mouse and navigating the game turned into fine motor practice and mouse familiarity that he wouldn't have got from a touchscreen app.
What this means
EdTech products are built for the average child. They have to be as the economics of software require reaching as many users as possible. My son is not the average child. He is a specific child who is obsessed with a specific game, sitting a specific test, at a specific moment in his learning. No commercial phonics app was going to be optimised for that and wouldn't bring the same engagement as something designed just for him.
What AI has changed is that the cost of building something tailored has dropped dramatically. This has implications beyond parents building games for their children. Tutors could build exercises around a student's specific interests. Teachers could adapt materials to a class rather than a curriculum. The constraint has always been time and technical access, not imagination, but that's changing with AI.
The code for the initial game came from a single conversation in Claude. You can find the code for the final version on GitHub and play it here.