OpenAI launches Codex: cloud agent to write and fix code

OpenAI has launched Codex, a new cloud-based software engineering agent that can handle multiple coding tasks simultaneously, according to a detailed announcement from the company that highlights its capabilities to transform how developers work. The AI-powered tool can write features, answer questions about codebases, fix bugs, and propose pull requests, with each task running in its own isolated cloud environment preloaded with the user's repository. As revealed by OpenAI, Codex is powered by codex-1, a version of their o3 model specifically optimized for software engineering tasks through reinforcement learning on real-world coding challenges.

Users can access Codex through the ChatGPT sidebar by typing prompts and clicking "Code" or "Ask" buttons to assign new tasks or questions, information shared by OpenAI shows that tasks typically take between 1 and 30 minutes depending on complexity. 

After completion, Codex commits its changes and provides evidence of its actions through terminal logs and test outputs, allowing developers to trace each step. The company notes that users can then review results, request revisions, open GitHub pull requests, or integrate changes into their local environment. OpenAI engineers have already incorporated Codex into their daily workflow, using it to handle repetitive tasks like refactoring and writing tests.

The Wall Street Journal reports that this launch represents OpenAI's significant push into the competitive AI coding domain, which has seen increasing investment from major tech companies like Microsoft, Amazon, Google, and Anthropic. According to the publication, OpenAI is looking to strengthen its position in this space where it doesn't yet have the same dominant status it enjoys with consumer-facing AI chatbots. The Journal reveals that Srinivas Narayanan, vice president of engineering at OpenAI, acknowledged the competitive landscape, stating "There are lots of good AI systems out there. Competition is clearly there."

Codex works differently from many existing coding tools that collaborate with developers in real-time, as The Wall Street Journal explains that the agent operates independently in the cloud and delivers output within one to thirty minutes. The publication notes that OpenAI claims the tool can better infer an organization's coding style and assist with code reviews compared to competitors. This represents "a fundamentally new way of working," according to Narayanan, who added that the company will collect feedback during this research preview phase to refine the technology further and mitigate potential risks.

The New York Times reports that OpenAI is also in talks to acquire Windsurf, another AI-powered programming tool, for approximately $3 billion, which would represent the company's largest acquisition to date. According to people familiar with the discussions cited by the Times, this potential deal could help OpenAI gain thousands of new customers in the tech industry, which is rapidly embracing tools that can instantly generate computer code. The newspaper explains that this acquisition would be part of OpenAI's broader strategy to expand its offerings beyond its popular ChatGPT chatbot and strengthen its position in the growing AI coding market.

Sources:

https://openai.com/index/introducing-codex/

https://www.wsj.com/articles/openai-launches-new-ai-coding-agent-c8dabc60 

https://www.nytimes.com/2025/05/16/technology/openai-codex-programming-tool.html