OpenAI debuts Codex: AI that codes and tests itself
OpenAI is launching a new cloud-based software engineering agent, Codex, as enthusiasm for using AI to build software continues to surge.
Designed with professional developers in mind, rather than hobbyist "vibe coders," Codex allows users to automate more of their coding workflows in a way that promises to be safer and more transparent than many existing tools.
Starting today, OpenAI’s Codex is available via the web for ChatGPT Pro users. The tool can not only generate lines of code but also navigate directories and execute commands inside a virtual computer—streamlining more of the tasks developers typically perform when writing code.
Just ahead of Google I/O
This move comes as OpenAI’s rivals Anthropic and Google have also introduced advanced coding-capable models. OpenAI’s announcement comes just ahead of Google’s anticipated unveiling of a more powerful coding tool at its I/O event next week, as reported by The Information. Additionally, several reports suggest OpenAI is in talks to acquire Windsurf (formerly Codeium), a startup known for its popular AI coding tool, in a deal valued at around $3 billion.
A major hurdle with "vibe coding" has been the tendency for AI-generated software to become opaque and harder for humans to understand and debug. OpenAI claims the Codex model has been specifically trained to explain its processes more clearly, assisting developers in identifying and fixing issues. By operating within a virtual computer, the system is designed to be inherently safer.
While tools like ChatGPT and other AI chatbots already allow users to write and analyze code, OpenAI has previously offered a Codex command-line tool for code generation. The newly released web-based Codex—labeled as "research preview"—takes things further by running its own mini computer inside the browser. This setup enables the agent to autonomously run commands, explore files and folders, and test its code.
According to OpenAI, this model reflects the company's vision for the future of software development, where agents will operate independently on their own virtual machines, handling tasks delegated to them by developers.
Well-known partners
Codex is already being used by several external companies, including Cisco, Temporal, Superhuman, and Kodiak.
The rise of "vibe coding" has been fueled by a new generation of AI models capable of writing and debugging code with remarkable proficiency. These same tools also help experienced developers accelerate their work.
Codex joins OpenAI’s growing portfolio of agentic AI tools, which includes Operator—an agent that controls a web browser to automate online tasks—and Deep Research, which performs extensive web searches and analysis to compile detailed reports.