Vibe coding: AI startup founder attempts to clone GTA in days

What happened?

Ziwen Xu, founder of the AI startup Hypercho, announced on May 12, 2025 on X (formerly Twitter) that he would attempt to create his own version of Grand Theft Auto 6 using Claude Max 20x, Anthropic's most powerful generative model available to the general public. The project, which he dubbed 'GTA VI Caliber', is based on the technique of 'vibe coding', which involves describing in natural language what you want to build and letting the AI generate the code.

According to Mashable, Xu started the project four days ago and has been posting daily updates on X, as well as sharing the code on GitHub. His stated goal is to 'beat the release of the real GTA 6', scheduled for late 2025. 'Ambitious, probably stupid, but I'll do it anyway,' he wrote.

Why is it important?

Xu's experiment tests the current limits of generative AI in creating complex software. While tools like GitHub Copilot or Cursor already assist human programmers, 'vibe coding' promises a qualitative leap: that AI understands high-level intentions and generates complete video games. Although it is unlikely that the result will compete with a Rockstar Games title (whose budget is estimated between $1 billion and $3 billion), the mere fact that a single person can sketch a functional game in days, rather than years, could democratize video game development.

Furthermore, the case illustrates the growing phenomenon of 'vibe coding', popularized by Andrej Karpathy, which advocates letting AI generate code from vague descriptions. This could lower the barrier to entry for creators without technical experience, but also raises quality, security, and intellectual property risks.

Consequences and risks

From a technical standpoint, Xu's project faces enormous obstacles. A game like GTA requires millions of lines of code, 3D modeling, physics, AI for NPCs, animations, sound, and a coherent narrative. Claude Max 20x, as powerful as it is, generates code snippets that must be manually integrated and debugged. The progress shown so far (oval characters moving in a basic environment) is far from a marketable game.

Legally, Rockstar Games and Take-Two Interactive are known for fiercely protecting their intellectual property. Although Xu claims his code is original, merely emulating the look or mechanics of GTA could lead to lawsuits for copyright or trademark infringement.

Economically, if 'vibe coding' matures, it could drastically reduce video game development costs, currently dominated by studios with hundreds of employees. However, it could also generate an avalanche of low-quality clones, saturating the market.

What should readers know?

Xu's experiment is, above all, a proof of concept. Don't expect a playable GTA in weeks. The developer community watches with interest but skepticism. The real question is whether generative AI can ever replace human creative work in producing high-quality interactive entertainment. For now, the answer is no.

For industry professionals, this case underscores the importance of adopting AI tools as assistants, not substitutes. For investors and startups, it suggests that 'vibe coding' could be a trend to watch, but with caution. And for gamers, it's a reminder that tech hype often outpaces reality.

“The goal: to beat the release of the real GTA 6. Ambitious, probably stupid, but I'll do it anyway.” — Ziwen Xu on X.

In summary, what happened is a bold but limited experiment that reflects the current state of generative AI: impressive in rapid prototyping, but unable to replicate the complexity of handcrafted masterpieces. The future of video game development will likely combine human creativity with AI efficiency, but we are still far from a machine writing the next GTA.