OpenAI Unveils Codex: Web-Based Agent Automates Coding Tasks
Updated on
Published on
OpenAI has rolled out Codex, a cloud-hosted coding agent now live for ChatGPT Pro subscribers. Running inside its own mini computer in the browser, the tool can hop through directories, execute shell commands, and generate or refactor code while narrating every step—an attempt to keep AI-written software transparent and fixable instead of a black box.
The release comes as “vibe coding” tools from Anthropic and Google race for mind-share, but Codex aims squarely at seasoned developers who want AI to handle grunt work without sacrificing safety. By sandboxing all actions in a virtual machine and explaining its logic line-by-line, OpenAI says Codex lowers the risk of hidden bugs and makes debugging far less painful.
Codex joins Operator and Deep Research in OpenAI’s growing suite of agentic helpers, hinting at a future where ChatGPT works more like a versatile coworker than a chat window. Early test partners—Cisco, Temporal, Superhuman, and Kodiak—report that the agent speeds up everything from API stitching to automated testing. If Codex delivers, the era of AI pair-programming just found its next gear.