Generative AI has a dirty secret. It lies. If you ask a language model to write a Python script, it might invent a library that does not exist.
If you ask an image generator for a car, it might give you five wheels. In a chat window, this is just a funny quirk. In game development, it is a fatal error.
Jabali AI is a San Francisco Bay Area company backed by a pre-seed round closed in early 2024 from BITKRAFT Ventures and the Sony Innovation Fund. They are tackling this specific failure point head-on. By integrating Anthropic’s Claude models into their core “Vibe Coding” engine, they are building the first AI engine designed purely for deterministic, reliable output.
The “Deterministic” Problem
Video games run on extremely rigid logic. If a character jumps, gravity must pull them down at the exact same speed every single time. The problem is that most AI models are designed to be creative and random. That randomness completely breaks games.
You have probably seen other vibe-coding tools that try to make games. They usually spit out frustrating, broken demos because the AI hallucinates the code. Jabali bypasses that broken demo phase entirely to give everyday creators fully playable, instantly shareable games.
Vatsal Bhardwaj is the founder of Jabali AI and a former executive at AWS and Meta. He realized early on that standard language models could not be trusted with game physics.
“You cannot build a simulation on a foundation of hallucinations,” says Bhardwaj. “We needed a model that understood code structure better than it understood poetry.”
This gives the engine the vital reasoning capabilities needed to ensure the math actually adds up when a user asks for a functioning economy.
Building Technical Credibility
For a new engine to succeed, it must actually work when a creator tries to use it. The biggest challenge for AI tools right now is proving they are not just vaporware. Jabali addresses this by using a highly intelligent hybrid approach.
The creative layer uses diffusion models to generate visual assets like textures and meshes. In this specific layer, a little hallucination is actually a good thing. It helps create highly unique and expressive art styles.
The logic layer operates completely differently. Every single line of code generated is cross-referenced against Jabali’s self-healing library.
This ensures the physics and economy math remain perfectly consistent across complex 3D environments. This separation of art and logic is exactly how Jabali helps creators hit their 48-hour concept-to-playable benchmark. It allows the engine to be incredibly expressive without ever breaking the game.
The “Open Intelligence” Stack
Proprietary engines usually lock you into a single ecosystem. Jabali is taking a different path by positioning itself as an open intelligence platform. This means it is model-agnostic, even though it is currently optimized for the intense reasoning power of Claude.
This architectural choice has really paid off. The engine’s ability to generate bug-free code was a huge driver in securing their $5 million early 2024 pre-seed round. Investors like the Sony Innovation Fund are not just looking for cool graphics. They are looking for scalable infrastructure that everyday creators can actually rely on.
The Bottom Line
The next generation of AI will not be measured by how creative it is. It will be measured entirely by how reliable it is. By solving the hallucination problem, Jabali isn’t just making games easier to dream up. They are making them genuinely safe to ship.
About Jabali AI
We built Jabali because we are tired of AI that hallucinates when it should be working. In game development, close enough simply isn’t good enough. If the physics or the logic breaks, the entire game breaks.
Headquartered in the San Francisco Bay Area and founded by Vatsal Bhardwaj, Jabali is an AI-native engine designed specifically for deterministic production. We are moving beyond the randomness of standard AI to give everyday creators a professional-grade environment. Here, gamers, modders, and indie dreamers can build deep 2D and 3D worlds using the natural language they use every day.
Our ecosystem is anchored by Jabali Studio. It acts as a self-healing bridge between your creative intent and functional code. We are not just building software. We are giving creators the tools to build things they never could before. Supported by a $5 million pre-seed round closed in early 2024 from BITKRAFT Ventures and the Sony Innovation Fund, we are proving that your creative taste is your most valuable asset.
Join the Movement:
Official Website: www.jabali.ai
GDC Game Jam Details: https://www.jabali.ai/announcement/jabali-gdc-gamejam-2026/
Technical Documentation: GitHub/Jabali-AI
Executive Leadership: Vatsal Bhardwaj on LinkedIn
LinkedIn: Jabali Official