COMING SOON: A New Way to Earn Passive Income with DeFi in 2025 LEARN MORE

OpenAI releases Codex, its flagship AI coding agent, in ChatGPT

In this post:

  • Open AI launched Codex, an artificial intelligence coding agent, on Friday. 
  • The tech firm revealed that the AI coding agent would need between 1 and 30 minutes to create simple features. 
  • Open AI said Codex could handle many tasks in parallel, such as fixing bugs, answering questions about the user’s codebase, and writing features.

OpenAI on Friday announced Codex, its latest flagship AI coding agent. The tech company stated that the AI coding agent would take between one and thirty minutes to write simple features. Each Codex task would run in its sandbox realm in the cloud, preloaded with the user’s repository.

OpenAI released the research preview of Codex, a cloud-based AI agent that could work on multiple tasks in parallel. The firm noted that the AI coding agent could suggest pull requests for review. Firms like Cisco and Temporal have already started using Codex in their operations as they provide feedback to OpenAI. 

Open AI reveals how Codex works

Per the announcement, Codex is supported by Codex-1, a version of the firm’s 03 designed for engineering tasks. According to OpenAI, the AI agent was trained utilizing reinforcement learning in various environments. It was trained on real-world coding tasks to generate code resembling human style and PR preferences. Codex was also trained to precisely adhere to coding instructions and run tests until it got a passing result. 

OpenAI said it launched Codex to users with ChatGPT Pro, Team, and Enterprise on Friday. The firm added that the AI coding agent will soon be available for its Plus and Edu subscriptions. The AI company advised ChatGPT users to access Codex through its sidebar. The firm added that users could assign new coding tasks to the AI coding agent by writing a prompt and tapping ‘Code.’ The tech firm said users could also ask Codex questions regarding their codebase. 

See also  Mark Zuckerberg bent over backwards for Trump, but got little in return

The AI firm said the new coding agent could read and edit files. Codex could also run commands like test harnesses, type checkers, and liters. The AI agent could be monitored in real time as it handles multiple tasks. 

The firm explained that once the AI coding agent completes a task, it would commit changes to its realm. OpenAI added that the AI coding agent would also offer verifiable evidence of its actions. The company explained that Codex would provide citations of test outputs and terminal logs, enabling users to trace each of the steps taken during the execution of the task. OpenAI said users could review the results offered by Codex. The AI firm added that users could also request revisions.

OpenAI praises Codex’s speed, security, and transparency

Josh Tobin, OpenAI’s Agents Research Lead, said the firm wanted their AI coding agents to act as online teammates who could execute tasks autonomously that could take human engineers hours or days to complete. The firm revealed it was already using its new AI coding agent internally to offload repetitive tasks, draft documentation, and scaffold new features. 

While announcing the new AI coding agent, OpenAI said it prioritized security when designing it. The firm also prioritized transparency when creating the AI coding agent. OpenAI clarified that they prioritized transparency so users could verify Codex’s outputs. Alexander Embiricos, OpenAI’s Product Lead, said the safety work and features on the firm’s o3 also applied to Codex. OpenAI stated that Codex would refuse requests to create malicious software. The tech firm added that Codex functions in an air-gapped realm, meaning it would not access external APIs or the broader internet. 

See also  Elon Musk quietly confirms X money "happening soon” amid dollar troubles

The firm advised users to review the code generated by Codex manually. The tech company added that they should expect Codex to communicate when faced with uncertainty or test failures. OpenAI said the Codex agent was designed to communicate issues to users so they could make informed decisions on how they intend to proceed. 

In recent months, Google and Microsoft revealed that roughly 30 percent of their code was created by AI. In February, Anthropic launched Claude Code, its own agentic coding tool. Google launched additional features for Gemini Code Assist, its AI coding agent.

Cryptopolitan Academy: Want to grow your money in 2025? Learn how to do it with DeFi in our upcoming webclass. Save Your Spot

Share link:

Disclaimer. The information provided is not trading advice. Cryptopolitan.com holds no liability for any investments made based on the information provided on this page. We strongly recommend independent research and/or consultation with a qualified professional before making any investment decisions.

Most read

Loading Most Read articles...

Stay on top of crypto news, get daily updates in your inbox

Editor's choice

Loading Editor's Choice articles...

- The Crypto newsletter that keeps you ahead -

Markets move fast.

We move faster.

Subscribe to Cryptopolitan Daily and get timely, sharp, and relevant crypto insights straight to your inbox.

Join now and
never miss a move.

Get in. Get the facts.
Get ahead.

Subscribe to CryptoPolitan