Vibe coding means building software by telling an AI what to write in plain language, and one founder is now posting daily results to see if the approach can ship a credible open world before Rockstar's GTA 6.
Two days after posting a blue oval hopping across gray blocks and labeling it "Day 1 of building GTA 6," Ziwen Xu posted a 3-second clip of a more human-shaped figure walking through what looked like an urban landscape. The visual leap was modest by game-development standards. It was also, by Xu's design, generated entirely through natural-language prompts to Anthropic's Claude, with no traditional code written by hand.
Xu, a 25-year-old AI startup founder, has framed the project as a public, day-by-day attempt to ship a credible open-world game before Rockstar's GTA 6 arrives in November 2026. The stakes he has set for himself are not the same as the stakes a reader should bring. The interesting question is not whether he beats Rockstar. It is what vibe coding actually produces when a single user, working in public, points it at the most anticipated entertainment product of the year.
Vibe coding is the practice of describing what you want in plain language and letting an AI assistant generate and debug the underlying code. The term was popularized by Andrej Karpathy, a former Tesla AI lead and OpenAI co-founder, and the inventor's public position on his own coinage has since soured. Karpathy has described his own daily use of AI coding agents as "net unhelpful," framing the tools as a productivity drain once the novelty fades, according to Gizmodo's reporting on the broader vibe-coding debate. On the other side of the same debate, another AI-coding founder has publicly proposed escalating a project like Xu's to a more capable Anthropic model, a Fable 5 or Mythos-tier release, to test whether the ceiling is the workflow or the model itself.
What Xu has actually produced, in the time since he upgraded to Claude's Max 20x plan, is a working prototype loop. The Day 1 artifact is a 3D blue oval that moves and jumps between gray blocks. The Day 2 clip adds a humanoid figure in a city-like environment. The project repo is public on GitHub, and Xu is posting fresh updates from his X account, as reported by Gizmodo's Bruce Gil. None of that is a game. It is, however, a continuous, dated progress feed that a reader can check against the marketing.
The structural gap remains the actual story. GTA 6 has been in development for roughly a decade, backed by a studio with thousands of employees, proprietary engines, and a track record of releasing open worlds that define a console generation. Xu's project, by his own description, is a one-person, prompt-only operation. The reasonable question is what would have to be true for the answer to Xu's framing to be yes. To get there, the public repo would have to show, in a verifiable way, a navigable city, traffic, pedestrians, a mission system, a save system, and the kind of asset density that makes a sandbox feel inhabited, and it would have to do that on a deadline that has already started to compress.
The repo is open. The updates are public. The two poles of the broader argument, the inventor of the term calling the practice a net negative and a competing founder proposing a much more capable model to push the same workflow, are both visible in the conversation Xu has inserted himself into. The next credible milestone is not a marketing post. It is a day-numbered artifact in the public repo that a reader can load and walk through. That is the metric this story will be measured by, and the project is, by design, the kind where the metric is the only thing that matters.