The journey
In 1982 I was 13, and I got a Dragon 32 for my birthday. It was my first computer. I loaded games from cassettes, and sometimes the tape snapped or got stuck in the recorder, and everything started again from the top. Later I bought a disk drive, and it felt like magic.
In October 2026 I started rebuilding it in the browser. I'm not a programmer; I build with AI. Here I write what happened, day by day: what we built, what I decided, and what I learned.
Newest first. To try it: the app · the Dragon machine · and what people read then: What Dragon Owners Were Reading
Testing everything, and looking for a real Dragon
Machine-code programs from 1982 called routines at fixed addresses inside the ROM. In our machine, about 60 of those addresses are now in their original places. Then we ran all 286 acceptance tests, one by one, and found no fault. What's left are 16 questions only a real computer can answer. A program you saved in the app now opens on the Dragon machine with one tap, and two pages were added: this journal and the history of the Dragon.
In the evening I published episode 2 of the journey. To answer those 16 questions I looked on eBay for a real Dragon 32. For now I've let it go.
What I learned: cheap isn't always cheap. The cheapest machine had no power supply, and a Dragon's is a special transformer, not just a cable. With one added, the price was almost that of a machine tested and working.
The whole BASIC in one chip
Our ROM is complete: the whole BASIC in one 16K chip, like the original, with the cartridge slot left empty. In went the cassette in the original format and speed, the printer, a RESET that keeps your program, and the 25 extension hooks that cartridges used to plug into BASIC. 574 tests passed, I wrote "merge", and that evening it went live.
What I learned: the AI suggested fixing two "bugs". Before fixing, we sent the reader to check the original. One was a real bug; the other turned out to be exactly what the original does, and "fixing" it would have taken us further from the original. The AI is confident even when it is guessing.
One chip, and Ciaran answers
I decided the whole BASIC must fit in one 16K chip, like the original, with no extension chip. The experiment showed it was possible, but tight: sixteen rounds of saving space, byte by byte.
I wrote to Ciaran Anscomb, who built the XRoar emulator, and he answered the same day: the exact cassette speed and other details no book has. The same day, a member of the Dragon owners' group told us that READA without a space didn't work for us. It did on the original, so we fixed it.
The principle
I stopped and wrote down the project's principle: the real computer, as close as we know how, without copying the original code. No shortcuts, however long it takes. Every decision is judged by one question: does this bring us closer to the original? Hence the new name, "the Dragon machine", and a work plan in stages that starts with research, not code.
The real computer
A long day. I shared a video of the computer in groups of old-computer fans, and the app got a new design, an English version, an accessibility page and a world Hall of Fame. In the evening "the real computer" was born: the XRoar emulator emulates the Dragon's chips, and on it runs our own BASIC, in the 6809's machine language.
When I checked, I saw that from the outside it looked like the original, but not on the inside. Programs were stored in memory differently, and the chip filled up until we added a second chip in the cartridge slot. I decided to rebuild: on the BASIC that Microsoft published as open source, and "identical" would include memory too.
What I learned: it's easy to get excited about something that looks right. The real test is what happens inside.
A home of its own
The Dragon moved out of my personal site into a home of its own, dragon1982.com: an app you can install on your phone, open source, and our own power-on message instead of Dragon Data's.
A green screen
It started as a small corner of the "evolution of the machine" page on my personal site: a green screen with a little BASIC we wrote in JavaScript. PRINT, GOTO, FOR, SOUND. By the end of that day there was a cassette with loading noise, "Guess the number", the first cartridge and a magazine program.
Behind the scenes
- Two rooms: the original ROM is under copyright. So one AI agent reads the original and reports only facts, in its own words. A second agent writes the code without ever seeing the original.
- Tests: every change passes hundreds of automatic tests. But tests only check what we knew to write, which is why people with a real machine matter.
- My rules: nothing goes live until I write "merge". No building before research. And nothing of others goes in without permission: no code, no games, no logos. The original ROM never goes in at all.
What's next
Machine-code games jump to fixed addresses inside the ROM. Our routines now sit at those same addresses, and 13 original games from the period already run on the machine without any change to their code: by Ken Kalish, Scott Adams and Mike Hughey, with their permission. Next: comparing the machine with a real Dragon, which I bought for exactly that. What we are building now · The history of the Dragon