OpenAI’s New Codex Agents Get Closer to Downsizing Your Dev Team

Openai only discovered The latest iteration of the Codexa AI agents with cloud movement that can simultaneously deal with more programming tasks without connecting your laptop’s resources. Announcement comes only one month after Openii posted for free, Open Kos Version, “Codex Cli.”

Unlike the usual tools to complete the code or generation of code we are used to during vibration coding sessions—What we repeated several times with AI chatbot until the results are satisfactory-codex acts as a semi-automatic agent with our own computer resources. He is able to handle everything from bug hunting to complex reaffactoring.

“S “Today we are going to take a step up to what we think is going on software engineering and publishing a new system, which is a remote software agent that can perform many tasks in parallel.”

Demo showed that open developers implemented more AI agents who work in parallel on different parts of the codes base, each in an isolated environment, doing different tasks in parallel with minimal human leadership.

“This change should probably take me at least 30 minutes or even hours to eliminate the removal of errors,” Hansen noted, one team member, during the presentation, while the AI ​​ended the task in the background during the interview. “We consider Codex as a reliable, if not more reliable than our own associates,” Katie, OPENAI researcher, claimed.

Unlike the previous Cli versionThis new Codex works on Openi’s infrastructure instead of locally. His agent architecture also means that AI can be examined and improved. In other words, developers can simultaneously release more of the encoding tasks and return later to review the results, as the agent can review their own outputs – and repair their own mistakes.

The system runs a new model called Codex Onewho described Openi as his “best encoding model to this day.”

“We took O3 and optimized it not only for reference values, but also for the type of code that people actually want to connect into their codes base,” Brockman said.

Openii emphasized that Codex is still a “research review” and said that there is more development ahead.

But you will have to pay your salary for the premium Openi layer you want to test it. Initially, Codex is available for Chatgpt forEnterprise users and teams, with plans to extend to plus and EDU users later in the future. Openii said he starts with “very generous rate limitations” and that there are no additional prices, although this will change because it collects information on use and feedback.

Looking in advance, openii plans to integrate Codex with trackers for issuing and CLI systems, potentially automating even more life cycle of development. The company also continues to develop a Codex Cli, an open -ender, a local agent who works on his own programmer machinery and predicts a future where local and distant versions work together.

“What you really want is a remote associate with your own computer, but who can look over your shoulder,” Brockman explained. “You typed there, work on some changes and like,” Ah, I want to go to lunch. Codex, can you finish this? “He simply takes it unobstructed and triggers it in the cloud.”

IN X postOPENAI CEO of Sam Altman said he had received very different reactions to the announcement.

“Alas, so far DM in shocking even dead heat between:” You made a software engineer and didn’t include an unlimited use of $ 20? Fuck! “” He wrote, “and” you made a software engineer and don’t charge $ 20,000 a month for that? What fuck? “

Edited Andrew Hayward