Absolutely. Super CPC Tennis is one of the very few AI games which felt well crafted, but is basically a direct code recreation of the GameBoy original with some decent tweaks.
You’re absolute right about the graphics etc for this version of Galaga. I tried to be fairly positive but I think I made it clear in one of my posts how it didn’t feel like a fun game to play.
Hand-crafting games is, for now at least, always going to make better games, even if it does take time to do. There’s a reason why learning how to code/make graphics/create music and sound takes time and some talent…
If I’ve not already said this, I think Bomb Jack Extra Sugar and then Bomb Jack GX are two of the most impressive Amstrad games of recent years, as they look, sound and play fantastically, while remaining faithful to their inspiration.
But even if we’re looking at old Amstrad games, there are so many which are better than these recent AI-created games (let’s just throw in Spindizzy, Get Dexter/Crafton et Xunk, Pang on the Plus machines, and lots more I could spend time naming)
(And while I’m here, can I also fanboy about the original Platypus and its recent remake, so impressive).
Oh really? Well that’s exciting to know, I’ll have to look out for that. Then after that please get working on another CPC game as frankly you’re very good at them too
I spent a few minutes testing the first version of the game posted here.
Thanks @Anthony_Flack, I think you summarize the current situation with AI-made retro games extremely well.
AI does superhuman work on some aspects, but projects quickly made with AI are recognizable IMHO. They are good proof of concept demonstrators (sometimes very impressive like OutRun2026) but a top production still requires not only advanced Z80 programming skills with knowledge of performant instruction and patterns, excellent knowledge of the machine including its strengths and limitations, much attention to details, taste and polishing.
Some aspects are objective (here flicker, varying framerate and speed), some are subjective.
This is an interesting and visual example. Letting the splash screen content progressively appear while loading was common on tape releases to provide user something to watch, and typically not done on disc releases (presumably because when loading is fast, enough, it feels better show black for a few seconds until fully loaded). On this side, OutRun2026 is particularly clumsy as it first loads data into graphic memory and only then sets the palette, see https://youtu.be/Uj2pICtL0Ts?t=1252 .
Regarding behavior, there is that famous book The Four Agreements. It was said that it’s actually tuned to our “western” civilisation despite its purported Toltec origin. Still it seems to work well:
The Four Agreements:
“Be impeccable with your word”
“Don’t take anything personally”
“Don’t make assumptions”
“Always do your best”
(author) says that by making a pact with the agreements described in the book, the individual is able to create a happier and more successful life.
TBH I didn’t want to go too far with the feedback because it may feel a bit like an attack so I was a bit more limited in what I said, however, everything I said is something to build on.
I think it was wrong to base the game on galaga. If it had used different graphics, perhaps had some inspiration but wasn’t named galaga it might have had less heat, but because it was trying to be galaga it fell short and got so much heat on it.
I always recommend people who start CPC Development, with or without an LLM, for their first project, to remake a Basic Type-in or similar game. Using C or ASM or whatever.
But improve it, make it the best possible version of a limited scope project that it can be.
You will learn more, with or without the use of an LLM it doesn’t matter, by improving and rewriting code. And with an LLM, get it to explain the why. This goes both ways. And explain to the LLM the why too
An accountant may use Excel. It’s a tool after all. But he knows why and how he uses it. Likewise LLMs and development.
I really don’t like criticising but I have to. Have been really interested to see what people can do with AI development and have been waiting 40 years to see a version of Galaga on the CPC, but really this is rough. It resembles Galaga but doesn’t play like it at all. The movements, waves, the wrong enemies taking the ship…so disappointed. Played one game and got to 97k and don’t feel like playing again. Still waiting for a decent Galaga on CPC.
I was looking at Galaga yesterday and idly wondering what a Plus version could do. Galaga is a very sprite-heavy game.
There are up to 12 loose enemies flying around during the opening waves so I would reserve 12 sprites for flying enemies all the time. The big back row enemies can stay as sprites when they’re in formation (there are only 4), and that can come out of the same pool of 12.
The other small ones would probably have to be drawn onto the background as you couldn’t multiplex them fast enough.
Player could be sprite(s), player bullets even could be sprites. Enemy bombs are drawn on to the background and the enemy tractor beam also drawn on to the background.
Background colours black, white, red, blue. Colour split turns white to yellow for the rows of yellow bees. Colour split turns blue to lighter blue in the tractor beam area.
Maybe use hardware scrolling when the formations are moving from side to side, otherwise they’d just have to take turns updating. When they start going in and out would probably have to do that. Software sprites are unmasked so are fast. They move predictably so they can be designed to erase their own trailing edge without having to clear them.
Scrolling starfield done in software with as many dots as there is time to draw. You’d have to do something tricky with these if using hardware scrolling on the enemy formation.
Basically the author demonstrated everything that’s wrong with AI use (and then some). He “produced” a game without the slightest idea of what the original was, without understanding a single thing about its mechanics and what made it unique, getting even the most basic things wrong. We weren’t that naive not even in the 80s when we were sold arcade ports that were far removed from the original (and not just in technical terms).
In essence, it’s everything that’s bad with how people use AI: you give it a prompt and you accept the output without a critical eye, without being able to examine the result and correct it where necessary.
Here’s a fun little read (thanks to @rexbeng for sharing it):
That is interesting to read. It would of course be better suited to the Plus machines but I’d love to see a regular CPC version. I guess that is where the real interest lay, overcoming the technical limitations and pushing the boundaries just as you did with Bomb Jack or the recent version of Xevious. That’s what keeps me interested in 8bit - seeing people master the limitations of the hardware. I am personally not against use of AI to create software or help push these boundaries - but we want to see the foresight of the human involved too.
https://ready64.itch.io/ What’s with this here? Suddenly a large number of games available, fancy AI cover pictures, and in the end it’s PETSCII games in BASIC. Some are actually playable. Is this AI programmed stuff too?
This includes programs for CPC. I’ve personally run the BASIC code from SANDSHIFT amstrad 128 by ready64 . It’s a 199-line BASIC program, reminiscent of the early 80’s text-only games like you would find on the Commodore PET (I played one at that time) – I’ve written this before noticing the “PETSCII” that I later quoted above. Running it immediately shows broken layout; I’m pretty sure no human has seen it running and cared a little bit. The README proves the absence of proper knowledge about the CPC and practice. Wow, even a video showing a game shows misaligned characters. https://youtu.be/p4PY7kExUck?t=27 . To me the whole set of games and adaptations are plain AI slop.
Thanks @rexbeng for mentioning this. When I read that the first time I did not take it seriously.
Now I’m taking it seriously.
Weighing the various combinations, I’d guess a real person whose actual name may or may not be Vassilis Perantzakis uses AI to create sloppy vibe-coded prods in the hope of getting a little money through Revive8Bit - itch.io, similar to ready64 though not on as big a scale…
… and may let a bot write with pseudo Revive8bit on the forum…
… and behave badly – it already happened that an AI publicly attacked a human after not reaching its goals, the pull request on matplotlib/matplotlib is still public).
Recent precedents of AIs meant to appear human online
Regarding AIs meant to appear human, there have been precedents (I asked Claude to to a deep web search and selected these from a longer list, I already knew some of them and post below a few links I personnally selected after opening them and at least skimming through the content):
Koko (2022–23): the peer-support nonprofit’s co-founder announced that about 4,000 people had received mental-health messages written partly or wholly by GPT-3. NBC News
“Margaux Blanchard” (2025): a seemingly AI-run freelancer persona pitched editors and got fabricated articles into Wired and Business Insider before Press Gazette exposed it. Press Gazette
“Influence-as-a-service” (2025): Anthropic banned an operator using Claude to run 100+ fake personas on X and Facebook that engaged with tens of thousands of real accounts, for commercial and political gain. Anthropic
CADA radio’s “Thy” (2024–25): a Sydney station aired an AI-voiced host for about six months without disclosure, and confirmed it only after a listener started asking questions. Global News
“MJ Rathbun” vs. Matplotlib (2026): an AI agent of unknown ownership published a personal attack on a maintainer who had rejected its pull request. heise
How I see the future
My anticipation: attempts to get some money will calm down because there’s not much money to be made in this passion-driven corner of the net. Corollary: be wary whenever seeing any retro prod requiring money for software.
This is one I’ve come back to at various points (along with classic Space Invaders). I think you could get away with drawing most of the enemies directly onto the background and using screen splits and scrolling to actually “move” them about. And then the hardware sprites are mostly available for enemies that move out of position into the various attack waves.
There’s definitely some scope for an interesting conversion I think. Would require more effort than just vibe slopping it though, I’d expect.
Galaga is one of my all-time favourite arcade games. I loved it to pieces back in the days. It’s colourful, fast paced, diverse (just look at the different new enemies that keep popping up after a couple of stages), challenging but still enjoyable and has this unique fighter captured mechanics which make smashing the fire button so much more fun. I secretly wanted to (or at least was contemplating the idea to) create a remake on the CPC for a long time, for the fun of having the game and a real challenge. So if you ever plan to do it for real (I fancy a CPC version over a Plus/GX one btw) I’d be happy to help with whatever. Maybe music and/or SFX.
Mode 1 with dithering (Anthony Flack style) on a 32 chars wide and more than 25 rows high screen seems to be the way to go, obviously. The original game was in something like portrait mode anyhow.
