Developer builds GTA VI clone with AI after losing patience with Rockstar's 13-year wait

After 13 years of waiting for Grand Theft Auto VI, 25-year-old developer Ziwen Xu stopped waiting and started building. Using Anthropic's Claude Max 20x plan, he created GT-Caliber - an AI-generated clone that produced playable environments in three days but also revealed the hard limits of current generative tools for game development.

Xu documented the project on social media over several days, detailing a workflow that leaned entirely on AI agents working around the clock. He used the tools to generate characters, buildings, vehicles, and interactive NPCs, all borrowing from GTA VI's visual style. The speed impressed onlookers. The errors did too - at one point, the system generated Los Angeles instead of Miami, the real-world basis for Vice City.

Engine-switching and the AI development loop

Xu initially built the project in the Godot engine, then pivoted after community feedback pushed him toward Unreal Engine. He tested multiple options before settling on Unity. The engine swap consumed an entire day of development, a reminder that AI code generation still requires significant human oversight when architectural decisions shift mid-project.

Despite the friction, the AI agents maintained momentum. Without direct human collaborators, Xu produced functional cars, weapons, and city environments in under a week. The generative code approach compressed what normally takes months into days, though the output remained a rough approximation rather than a polished product. Fans responded with enthusiasm to the rapid progress, sharing the project widely.

Where the AI stumbled

The Los Angeles-for-Miami error was not a minor glitch. It exposed a fundamental weakness in how generative models interpret prompts - the system understood "American city with beaches and crime" but missed the specific cultural and architectural markers that define Vice City. For developers watching the experiment, this underscored the gap between generating assets and building a coherent creative vision.

Xu's experience also highlighted tooling friction. The engine migration, the prompt-tuning required to get usable outputs, and the need to manually correct AI misinterpretations all added labor that the "AI builds everything" narrative tends to gloss over.

Rockstar's stance on AI in development

Take-Two CEO Strauss Zelnick has publicly dismissed the idea that AI can produce titles on GTA VI's scale. He said the technology can be a helpful tool but "is incapable of producing titles on the scale of GTA VI." Zelnick maintains that human talent and deep entertainment industry knowledge remain irreplaceable - a position that frames Xu's experiment as a proof of concept rather than a production-ready pipeline.

Rockstar has reportedly avoided AI during GTA VI's development, a deliberate choice to sidestep the controversies that have followed other studios and to protect the game's commercial trajectory. The contrast between the studio's approach and Xu's solo AI sprint illustrates the current divide in game development philosophy.

Why this matters for IT and development professionals

GT-Caliber is not a threat to Rockstar's blockbuster, but it is a clear signal about what small teams can now attempt. A single developer with a high-tier AI subscription produced a functional open-world prototype in under a week - something that would have required a team and months of work a decade ago. The experiment also surfaces the real bottlenecks: prompt specificity, engine compatibility, and the manual labor of correcting AI outputs. For developers evaluating Claude or similar tools in their own pipelines, the takeaway is that speed gains are real, but architectural decisions and quality control still demand human judgment. The tool accelerates execution; it does not replace direction.