OpenAI Shares Codex Results: 30,000 Visitors in 4 Days, and the Comments Are Already Doing the Math

The OpenAI Developers account posted a tweet. The message was direct: a creator named laraavci_ used Codex to build websites and mini-games, serving European tech entrepreneurs, bringing in more than 30,000 visitors in under 4 days.

The video shows what she built. One individual creator, plus an AI coding assistant, turned an idea into a product with real traffic within a few days.

Codex is OpenAI's AI coding agent, able to write code, debug, and deploy based on natural-language instructions. For people without a development background, it lowers the barrier to building a product by a huge margin. For people with development experience, it feels more like a partner that can keep working continuously.

The tweet itself is a standard product case study. But as you scroll down through the replies, the tone starts to split.

Some praised it. @vexor4i said this kind of growth is impressive and that they love seeing builders ship quickly and find traction.

Someone raised a question no one answered. @Krrish asked: How many people come back after day 4?

Someone focused on cost. @foundrceo said, with 30,000 visitors over 4 days, the hosting bill is the real boss fight.

Someone mocked the European startup environment. @Lincoln said this tech guide shipped faster than a European founder can open a bank account.

Others ignored the case study entirely and started demanding releases and criticizing the product. @Hakamite said if 6.2 were released, she would get better support. @VM said OpenAI does nothing but PR. @omer khan posted two screenshots, complaining that ChatGPT had been randomly changing things on its own for 12 days straight, wasting time and Codex tokens.

What started as an official showcase post turned the comments section into a live product-feedback session.

One reply deserves to be pulled out separately. @ShadowAguy said getting Codex to build the game is the easy part; 30,000 visitors in four days is the exam that an agent can never get graded on.

That reply takes apart the case study's narrative. AI tools can help you build something; that capability has been proven over and over. But what happens after you build it? Traffic arrives—what is retention? Can the servers handle it? On day five, is anyone still opening it?

OpenAI showed the first half. No one mentioned the second half.

It makes no sense to deny Codex's value. For one person to build something with 30,000 visitors in a few days would have required a small team three years ago. The barrier really has come down. But once the barrier is lower, competition gets more crowded. When everyone can ship quickly, shipping itself is no longer a moat.

Perhaps the most direct reply under that tweet came from @arror: one person plus an AI coding assistant can turn an idea into a product with real traffic in a few days. But when real maintenance is required, how long these quickly built things can hold up is an open question.

So what is really worth caring about in this case is not the number 30,000. It is day five.