Making games with human soul in the age of AI.🇻🇦Matthew 28:20

North Carolina, USA
Joined August 2016
Here’s the hard truth: There is no known meta for indie game dev with the rise of AI. Nobody knows what will end up being the smart play. Everyone’s experimenting. I, for one, find that incredibly exciting. So, these are my personal guiding principles going into 2027. 👇
1
2
2
2,131
Just realized I will never have to bake a lightmap by hand ever again
123
Introducing my humble model, QURSE-1-0.2b! QURSE is a classifier head paired with the Qwen3-ASR-0.6b audio encoder. It catches stumbles and repetitions in tongue twisters 99.57% of the time on just 1.8GB of RAM, trained locally. Details in🧵 Weights are available on HF. huggingface.co/woollyone/QUR… Play the tongue twister game here: woolly.one/qurse-demo
1
181
Performance: In my testing, Qwen and QURSE achieved 99.57% accuracy in detecting the stumbles while not falsely flagging valid speech variance. Crucially, it also runs on roughly 1.8GB of RAM, and thanks to the ONNYX architecture, relatively platform-independent (the WebGPU demo ran on my M2 Pro Mac and my NVIDIA RTX 5080 Windows rig. This is important because I wanted it to run on as low of a footprint as possible on most players' platforms.
1
35
Caveats and notes: - My use-case was narrow, so I chose to synthetically generate all my training data. I found pre-existing datasets to be insufficient. For instance, the stuttering dataset SEP-28k had mislabels, multiple speakers, and incomplete sentences. The tradeoff is I don't have great authentic human speech evals (yet). - The model runs locally I haven't tested it on a wide variety of architectures. - Weights are open, but non-commercial for now. When I release my full game, I will de-restrict the license. However, I'm not making any guarantees this will work for many other use cases. Like I said, it's a proof of concept!
14
I did my best to get some traction on my flying roguelike demo. I reached out to content creators, promoted on socials… …and I’m sitting around 38 plays, 0 ratings and 0 comments. Safe to say it’s not getting the market validation I need to bring it to Steam, so that likely brings us to the end of this game’s story.
Hopshot: Arenas is a "hare-raising" arena roguelike. - Zip and chain hits as a flying bunny - Upgrade stats and purchase absurd ranged weapons to keep up with the increasing madness - Fight a snake boss at in arena 10 #godot #indiegame #roguelike
1
53
Too bad! The game is definitely fun and the full version would have been super epic. But I learned a lot, and I have plenty of prototypes to validate from here. My next one in particular is going to be kind of insane.
31
It’s sad to see so many people in this industry losing their heads over this. This sloppery slope assumes AI will autonomously make all the games for people in the future with no soul. Either the human soul has some intrinsic value in art or it doesn’t; can’t have it both ways.
AI “vibe coding” is starting to create a serious problem for the video game industry. The release of Claude Opus 5.5 has made it much easier to decompile and recompile games, leading to a wave of unofficial ports, mods and game clones. Publishers are already filing DMCA takedowns over some of this activity. Kotaku Former Take-Two AI chief Luke Dicken says developers and publishers need to take the trend seriously: “This new wave of decompile-and-port is definitely something that developers and publishers need to take seriously.” “More invasive DRM, platform-level solutions, and probably more of a push towards streaming solutions.” Godot executive director Emilio Coppola puts more responsibility on platforms such as Steam, arguing that developers have very few ways to protect themselves while bad actors can profit from selling AI-made clones: “The burden should fall onto the platforms who are enabling this sort of behavior.” There’s also a growing concern among indie developers about showing unfinished games publicly. Coppola says he has heard from developers who are now: “Worried and anxious to even share early access versions of their games”
55
This is my first time using Gemini 3.8 Flash and it has a lot of those “H1 LLM” smells. The GPT-5.5 “oh wait is that what you wanted me to do?” cluelessness and lack of alignment. I’m sure 4 Flash will fix that but it’s weird to see now; it’s like a relic from a bygone era.
64
Thoughts on Gemini 4 Argon so far (I have a $20/mo plan): 1. It’s extremely well-aligned to game dev and game design. Some of the best taste that I’ve seen from a model; up there with Opus 5.5. 2. Two prompts blew through 27% of my weekly usage.
1
2
215
What the… Gemini 4 Argon is INSANE for game dev I asked it to rig my witch character and this is what it did
47
Can’t steal my game if I don’t even know what I’m building
17
"If an AI model dropped that was an order of magnitude better than the last, would you company fare better or worse?" is living rent free in my head lol
16
Robert Varadan retweeted
There’s something beautiful about the most expensive game of all time ($2B) being dropped the same month that AI models became capable of creating comparable experiences. The swan song for the Big Studio Era.
621
455
156
14,077
994,662
Hopshot: Arenas is a "hare-raising" arena roguelike. - Zip and chain hits as a flying bunny - Upgrade stats and purchase absurd ranged weapons to keep up with the increasing madness - Fight a snake boss at in arena 10 #godot #indiegame #roguelike
1
1
2
187
3. Taste is the moat I’m hearing a lot about how Opus 5.5 has great “taste.” What it really has is great alignment. The feeling of “yes, this is exactly what I wanted!” is highly optimized. It’s not coming at that realization itself, it doesn’t know; it’s just able to get you to that point faster than before. And so, if everyone has that ability, taste is the moat. You need to know what is worth putting effort into and what is a waste of your time and energy. What will hit and what will not. This either comes with experience or is present innately; but look around, it is in short supply. Now, taste is not a new moat, mind you. But since the effort required to bring about a game has diminished, it’s the most important moat.
1
96
This is the bet I’m making. It’s a bit optimistic, and maybe I’m wrong, but it’s a hell of a lot better than despair! And again, these are uncertain times. So get back out there and build with me!
83