Scenes From a Vibe-Coded Safari
So yes, I've fallen victim to the game vibe-coding spell. I ended up knee-deep in a project that started as a summer holiday toy and took me to the edge of what's possible with models like Fable and 5.6 Sol.
I've come back from this safari convinced that this is extremely feasible. A reasonable timeline says two years until all of hypercasual and hybrid-casual is vibe-coded, then another two until, say, 70% of the top 100 mobile puzzle games use it extensively.
If you've been on Instagram lately, around 70% to 80% of the game ads in my feed now appear to be AI-generated or, at the very least, AI-assisted. But progress has not been evenly distributed.
3D is progressing - see Tripo 3D - but less quickly than you might expect. BlenderMCP only arrived in March 2025. The models do not have an intuitive understanding of 3D space, so they are constantly taking pictures, analyzing them, and adjusting the scene. Getting the right model and texture is a painstaking process that just is not feasible for super-tight teams without external help. This will be solved. I have no doubt.
But for the other stuff - 2D art, internal tooling, BI, internal documentation, engine code, UX - AI is already here. With each model, the bar moves out, and end-to-end publishing is now feasible for a non-technical solo developer.
The limits exist, but genre plays a large role in fit. Early 3D, networking, OTA content delivery, and some of the finer points of content scale are tough without a smarter human architecture and someone who knows how to extract the right abstractions from the model. Ironically enough, AI is most powerful in a CTO's hands.
Anthropic regularly shares how it employs Claude internally, and its pattern is a perfect setup for games at today's multi-model capability levels. But it is still AI doing game development rather than game development using AI.
It is again obvious that we need an AI-native game development engine. Unity has moved into in-editor AI, but this still feels like an existing engine with AI bolted on. Unreal has been quieter. Neither amounts to the serious attempt the category needs, and that feels gobsmacking.
The reason VCs are not pouncing is that the business model does not appear attractive enough to make a new entrant profitable. Epic does not disclose Unreal's economics, but the engine has never looked like its core money maker. At Unity, Create Solutions revenue was less than half Grow Solutions revenue in Q4 2025.
So the tools are here, the cost is a $200 monthly subscription, and letting go of any resemblance to leisure, while the engine everyone needs is a business nobody wants to be in.