To AI or not to AI.
Posted: Thu Oct 08, 2026 8:11 pm
Most people seem to have concentrated on transcoding or porting games, or authoring games with AI from the ground up.
I wanted to do a test on the creation of something that fixed a perceived problem on the Dragon, having previously only used AI for design and reverse engineering, plus writing little tools.
The thing that always got me when I originally had the Dragon as a kid was the relatively limited sound capability.
Obviously then I knew of lots of 4 part stuff, and these days I'm familiar with the standard selfmod 4 voice DAC setup. I had a number of well formed ideas to try and extend that with an enveloped and toned percussive sound channel, and optional envelopes on the main voices. Mechanisms already understood, but not built and cycle counted to check genuine viability. Cost - a bigger engine, some noise, and larger data files, but hopefully ok. The main idea was to be able to have the Dragon be able to play a pop song, and have an engine that was capable enough to produce a fair shot at a decent chiptune style output.
Firstly - prove it's really OK - with AI build up a web-based emulator with demo patterns / tunes. Modify that through multiple iterations to prove plausibility. Make the demos little test cases for features, noise and perceptual optimisations etc.
Then plan out implementation in detail. Or just tell Claude to get on with it. Note, the code here has not been reviewed by a human. I will probably look at it at some point.
The result is attached. Unzip and load into Xroar on a Dragon 32 or 64 with the Load menu. EXEC 3400. Likely to work on a Coco 1 and 2 as well. 32k required.
I wanted to do a test on the creation of something that fixed a perceived problem on the Dragon, having previously only used AI for design and reverse engineering, plus writing little tools.
The thing that always got me when I originally had the Dragon as a kid was the relatively limited sound capability.
Obviously then I knew of lots of 4 part stuff, and these days I'm familiar with the standard selfmod 4 voice DAC setup. I had a number of well formed ideas to try and extend that with an enveloped and toned percussive sound channel, and optional envelopes on the main voices. Mechanisms already understood, but not built and cycle counted to check genuine viability. Cost - a bigger engine, some noise, and larger data files, but hopefully ok. The main idea was to be able to have the Dragon be able to play a pop song, and have an engine that was capable enough to produce a fair shot at a decent chiptune style output.
Firstly - prove it's really OK - with AI build up a web-based emulator with demo patterns / tunes. Modify that through multiple iterations to prove plausibility. Make the demos little test cases for features, noise and perceptual optimisations etc.
Then plan out implementation in detail. Or just tell Claude to get on with it. Note, the code here has not been reviewed by a human. I will probably look at it at some point.
The result is attached. Unzip and load into Xroar on a Dragon 32 or 64 with the Load menu. EXEC 3400. Likely to work on a Coco 1 and 2 as well. 32k required.