(Bloomberg) — OpenAI is rolling out a new artificial intelligence agent for ChatGPT users that’s designed to help streamline software development as the company pushes into a crowded market of startups and large tech firms offering AI tools for coders.
Most Read from Bloomberg
The agent, called Codex, will be able to write software features, fix bugs and run tests, the company said in a blog post Friday. Codex, which is still in the early stages and has limited functionality, is geared towards workers with some technical knowledge and will first be released as a “research preview” to paid ChatGPT Pro, Enterprise and Team users.
A growing number of tech companies, including Microsoft Corp.-owned Github, Alphabet Inc.’s Google and Anthropic, offer AI tools for programmers. Some startups, including Cursor maker Anysphere and Windsurf, have also attracted users and investors with AI-infused coding assistants that can analyze a software developer’s actions and suggest the next few lines.
In a sign of how important this emerging market is to the company, OpenAI is in talks to buy Windsurf for about $3 billion, Bloomberg News has reported. The deal would be the company’s largest acquisition to date.
AI agents are billed as tools that can field more complex requests on behalf of users with minimal supervision. OpenAI said its technical staff are already using the coding agent daily for a range of work, from repetitive tasks to helping build new features. Other companies, including Cisco Systems Inc. and Kodiak Robotics, have also been using the tool, OpenAI said.
“We’re just seeing very fast progress in the model’s ability to solve coding and software engineering problems,” said Josh Tobin, research lead on agents at OpenAI. “We see this as a particularly fast way for us to get to that agents vision.”
Codex runs on a version of OpenAI’s o3 AI reasoning model that is optimized for software engineering. The tool can take anywhere from one to 30 minutes to complete a task, depending on complexity.
OpenAI also said Codex was trained to identify and refuse requests aimed at the development of malicious software, a nod to concerns that bad actors could turn to more sophisticated coding agents for cyber attacks and other harmful uses.
Most Read from Bloomberg Businessweek