The Guy Vibe-Coding a GTA 6 Clone With AI Is Now Buying a PS5 to Play the Real One

The most honest thing anyone has said about AI and Grand Theft Auto VI came from the one person actively trying to copy it. Ziwen Xu spent the summer vibe-coding a GTA 6 clone with a loop of AI agents, racing to ship something akin to Grand Theft Auto before Rockstar Games released the upcoming sixthquel on November 19. Months later, his latest move is to tell everyone that he's buying a PlayStation 5 so he can go play and stream it.

We're not knocking on Xu though and it's not like he has already given up. If anything, he's doubled down on his previous project, except he's doing it differently now.

For those unaware, Xu, starting in June, pointed a stack of AI agents at the problem, posted the results daily, and made real headway. A blank plain became a blue oval, then a character, then NPCs, cars, driving, shooting, and a working menu that in his words "boots like a AAA game." He swapped engines from Godot to Unreal when the community told him to get serious, and he actually created something quite fascinating to watch for a couple of days.

Unfortunately, the project hit a wall that AI just couldn't sidestep.

You see, by using GTA as the base of the game he's trying to create, Xu found himself in legal hot waters. Months after starting the entire project, he tore the whole thing down in August, rebuilding it instead as an original game called Caliber, or CALI, with its own world and characters.

As it turns out, you can't actually make a GTA 6 clone because what makes GTA what it is belongs to Rockstar.

The other issues he faced had nothing to do with lawyers. Early on, the agents cheerfully built a skyline of Los Angeles skyscrapers when the goal was a Florida, Miami-flavored setting, because an AI doesn't have the creativity and ingenuity to create a Vice City that looks like, well, Vice City. Later, with the world scaled up, the frame rate dropped to roughly one frame per second, and he had to spin up yet another AI tool to try to claw it back.

Now, does this mean AI is useless because you have to use AI to solve a problem AI created? No. CALI exists and will continue to be for at least the foreseeable future. A year ago no solo developer gets this far this fast on prompts alone, but the reality is that an agent loop still can't make a game like GTA 6 and it will probably never happen.

This isn't an AI problem. No other human or studio can make something like GTA 6. It's thirteen years of dedicated work, poured into a single project, with a dedicated Los Angeles studio whose entire job is art-directing individual pedestrians, more than 600,000 unique NPC animations, over 50 real street artists hired to paint the murals, and actual race-car drivers tuning the handling. Digital Foundry called the result of this level of obsessive dedication the most impressive real-time rendering it has ever seen.

You see, an AI agent, however tireless, cannot create something original. It can imitate a building, but it can't decide why the building has to go there, or what a corner of Leonida is supposed to make you feel when you drive past it at dusk.

Humans can give you that, which is why GTA 6 uses no generative at all. Rockstar has made it clear that this is a handcrafted game, maybe the last AAA title built entirely by human hands. So the summer handed us a perfect natural experiment, one game built entirely by humans over more than a decade, and one built entirely by AI agents over a few months, aimed at achieving the same goals. Only one of them is in the midst of a historic marketing run, topping Netflix and selling out consoles that just went on a price hike.

When even the guy betting on AI wants to play the one made by people, you have your answer. AI can help build a game, but it can't just dream it up.