Katy Shi, a researcher who works on Codex’s conduct at OpenAI, says that whereas some people describe its default character as “dry bread,” many have come to understand its much less sycophantic type. “A variety of engineering work is about with the ability to take important suggestions with out decoding it as imply,” Shi says.
A number of main enterprises have signed on to make use of Codex too. “The truth that ChatGPT is synonymous with AI provides us an enormous benefit within the B2B market,” says Fidji Simo, OpenAI’s CEO of purposes. “Corporations wish to use applied sciences their staff are already accustomed to.” OpenAI’s technique to promote Codex is basically based mostly on packaging it in with ChatGPT and different OpenAI merchandise, Simo mentioned.
Cisco’s president and chief product officer, Jeetu Patel, says he has instructed staff to not fear about the price of utilizing Codex, as a result of they’ll have to be comfy with the software. When staff ask if “they’re going to lose their job as a result of they’re utilizing these instruments,” Patel says, “what we’ve to inform our individuals is not any, however I assure you may lose your job if you happen to do not use them, since you will not be related. So you are going to be out.”
Right this moment, the panic round AI coding brokers has unfold far past Silicon Valley. The Wall Avenue Journal credited Claude Code with inflicting a $1 trillion tech inventory sell-off final month, as traders feared that software program would quickly change into totally out of date. Weeks later, IBM’s inventory had its worst day in 25 years after Anthropic introduced that Claude Code could possibly be used to modernize legacy programs that run COBOL, widespread on IBM machines. OpenAI has labored tirelessly to make its AI coding agent a part of the societal dialog, spending thousands and thousands of {dollars} on a Tremendous Bowl business about Codex, somewhat than ChatGPT.
On the Mission Bay temple, nobody must be pitched on Codex. Many OpenAI engineers I spoke with mentioned they not often kind out code in any respect anymore. They only spend their days chatting with Codex. And typically they get collectively and do it in congregation.
At headquarters, I sat in on a Codex hackathon—about 100 engineers crowded into a big room. Everybody had 4 hours to construct the most effective demo with Codex. A senior OpenAI chief stood on the entrance of the room, twisting away from the laptop computer in his palms and talking crew names right into a microphone. Group representatives nervously walked to a podium and gave brief speeches about their AI initiatives by shaky voices. Winners acquired Patagonia backpacks.
Lots of the initiatives have been each created with Codex and designed to assist engineers use Codex higher. One group constructed a software that summarizes Slack messages into weekly stories. One other group constructed an AI-generated Wikipedia-style information to inner OpenAI providers. Many of those demonstrations would have taken days or even weeks to spin up beforehand, however now they are often finished in a day.
On my means out the door, I bumped into Kevin Weil, the previous Instagram government who’s now heading OpenAI for Science, the corporate’s new unit constructing AI merchandise for researchers. He instructed me Codex was engaged on some initiatives for him in a single day, and he would examine on them within the morning. That’s change into common apply for Weil, and a whole bunch of different staff. One among OpenAI’s objectives for 2026 is to develop an automatic intern that does analysis on (what else?) AI.

