Developer uses AI agents to build GTA 6-style clone

An independent developer is assembling a fan-made Grand Theft Auto 6-style game using multiple AI agents to generate environments, mission scripts and nonplayer character behavior. The creator is sharing code snippets, short gameplay clips and technical notes as an experiment in rapid prototyping.

The workflow, often called “vibe-coding,” assigns high-level tasks to several AI agents that iterate on code, art and dialogue with limited direct oversight. The project uses generative models that can write code, create 3D assets and textures, synthesize voice lines and draft simple mission logic.

Development is organized into modules such as world generation, NPC systems, vehicle physics and mission scripting. Agents propose implementations and refine them over repeated cycles, enabling a single developer or a very small team to assemble components that previously required larger groups and longer schedules.

Progress has been posted online through public code repositories, short video clips and technical write-ups. The developer frames the effort as a technical exercise to test how generative AI can accelerate open-world prototyping rather than as an attempt to replace a commercial release.

A website hosting project updates includes a notice that it is not affiliated with Rockstar Games or Take-Two Interactive and states: “‘Grand Theft Auto’ and the ‘Grand Theft Auto’ logos are trademarks of Take-Two Interactive.” That notice underscores intellectual property rights that apply to fan recreations and the limits on distribution and monetization.

Recreating the scope and polish of a full Grand Theft Auto title involves challenges beyond initial prototyping. Open-world games require extensive systems for animation, physics, AI behavior, networking and a large catalog of art and sound assets. Generative agents can speed concept art, scaffold code and draft dialogue, but human oversight is needed to verify stability, performance and consistency. Proprietary studio technologies and licensed assets are not available to public projects, which limits how closely a fan build can match a commercial release.

The project is part of a broader activity among hobbyists and small teams using agent-driven workflows and large models to automate programming tasks, produce placeholder or final assets and simulate behaviors. These experiments create early-stage demonstrations and prompt questions about ownership, attribution and acceptable reuse.

The developer documents the workflow, model prompts and integration challenges for others interested in AI-assisted game creation.