TL;DR: long post to tell my long term journey of a life of CPC prods, and if AI is used in the future it’s because I use AI as complements so that my human-conceived and hard-won prods are better, not lazy prompt-and-publish.
A long term project
Just to clarify a thing: I have had a project for a game for more than ten years.
Many steps to pave the way
There’s even a verifiable hint about it. When I first subscribed to the CPCWiki forum my username was “FindYWay”. It’s a hint about the prod I had in mind at the time. The prod evolved. I needed many steps along the way, actually in view for that magnum game.
In ~2013 there was no CPC development integrated environment that would build all tools from source, so I wrote cpc-dev-tool-chain. Later it was forked into Cpctelera (and about ten years after that I understood the spanish pun behind its name).
In ~2014-2017 I had to validate that non-trivial tasks could be programmed in pretty standard C code and run well on the CPC so I wrote color-flood-for-amstrad-cpc, it was visually not pretty but it validated the concept.
In 2020 I needed some experience about something higher performance, a game with music, very polished, with guaranteed 50fps animation on an ambitious area of the screen, I wrote Just Get 9 – still using the firmware for non-performance-sensitive stuff and keyboard scanning, which provided debouncing and buffering for free! Thank you everyone for your reception of this game which apparently established my reputation?
I also wanted to explore extremely high performance graphical CPC prod and wrote and released in 2025 Sunny Day which also was very well received, thanks.
In 2025 I wanted to emit a signal that I’d probably need the help of an artist that can create graphics for the CPC, so I made Want Graf released at Benediction Coding Party #5 Graphics competition. Is the use of AI a good idea to emit this signal, I don’t know. Anyway I thought with my own brain until I knew exactly the image I wanted, then had AI generate a base image, then used traditional tools like gimp to turn that into a proper MODE 1 image and transform that monstrosity computer into something that actually looks like a CPC6128, while manually keeping that dithering. Yes, clicking on individual pixels to get the dither right. Not at all prompt-and-publish.
Current status of the game
Though the initial idea predates all this, I’ve been actually working on this most ambitious game since, let me check, the first commit in the repo is dated July 2020, that makes 6 years. So far it has:
- some design documentation about a back story, the parts and their gameplay,
- a nice graphical intro screen,
- some extremely hand-optimized Z80 code for a specialized graphic routine (I’m not saying what it’s doing now) which currently handles one case and have to handle many others to be considered completed,
- some mathematical study that happens to be important for the game,
- also some Python code to check the total correctness of the algorithms I’m considering so that I’m certain of their rounding behavior given the register size and other parameters I’m planning, before I code them in Z80 assembly.
Okay, where is this going?
I’ve been working on this new game (mostly by small burst on holidays) for the past 6 years. So far I’ve not used AI in any part of it.
Is AI okay as a tool to write this game I have in mind?
The project is in my mind. I might use AI to save me time transferring this precise vision into a working game. AI will not drive the project. Or even a part. I may use AI as an accelerator. AI may allow me to finish the prod in less time or reach a level of completion/quality that would not have been possible without it.
Still the game will be a prod desired by a human, made by a human. 
The idea I have is precise because I have been carefully evaluating a number of decisions that make the game possible.
I won’t let AI drive the project.
Similar to what @BSC wrote, I expect all the Z80 code that matters to be written by me (non-performance-critical code would be written in C). I might ask AI to write a draft Z80 routine, but from my experience of having very carefully written the tile fall routine in Just Get 9 using all addressable Z80 registers (even the 8-bit halves of IX and IY, just not I or R) for speed, at this point I’m pretty sure AI won’t write code that satisfies my level of demand for Z80 speed and control.
When it’s released
Because of its polished ideas, intro, visual style, gameplay, original world and all, I hope when you see the game you’ll think “Wow 
”, then “I wouldn’t have even dared imagine a game like this was even possible on the CPC, let alone try to write it.” and pick your jaw from the floor. 
Long story short, if AI helps me finish writing the game I elaborated for years, I’m okay to use AI.
I converted it to cart so I could run it on my GX4000 rather than just emulating it on RVM and it looks really nice on a CRT (I don’t have an Amstrad monitor unfortunately). I’ll upload the CPR file to the conversions topic in a minute.
