OpenAI's AI Just Coded Non-Stop for 25 Hours Straight and the Results Are Terrifying

OpenAI just showed off something that should make every software developer a little nervous.

Their latest AI coding tool, GPT-5.3-Codex, was given a blank project and told to build something complex. It coded for 25 hours straight without stopping. No coffee breaks. No Stack Overflow. No asking a coworker for help. Just pure, continuous software development.

To understand why this matters, think about how software normally gets built. A human developer might spend a week planning, writing code, testing it, fixing bugs, and repeating that cycle. This AI compressed all of that into about a day.

This isn't just a party trick. OpenAI published the full details in a developer cookbook, showing exactly how their system handled "long-horizon tasks" (tech speak for big, complicated projects that require sustained focus).

This news comes at a particularly sensitive time. The stock market just crashed partly because investors are worried about AI replacing white-collar workers. And here's OpenAI basically proving those fears aren't crazy by showing their AI can out-work human programmers on marathon coding sessions.

The AI coding revolution is already well underway. Anthropic's Claude Code and OpenAI's Codex have surged in popularity among developers in recent months, and complex software that used to take weeks can now be built in minutes.

The question isn't whether AI will change software development. It's whether there will be enough software jobs left when it does.

As reported by OpenAI and Radical Data Science.

Source: OpenAI