The daily work of building AI has been transformed by the very technology being built. At OpenAI, most technical staff now use Codex - the company's agent-based coding tool - and the shift has been dramatic enough to prompt chief scientist Jakub Pachocki to declare a new reality.

"There's a big change happening, especially in programming," Pachocki told MIT Technology Review in an exclusive interview. "Our jobs are now totally different than they were even a year ago. Nobody really edits code all the time anymore. Instead, you manage a group of Codex agents." "Hvylya" reports on the key takeaways from the conversation.

The admission is striking coming from Pachocki, who described himself as a purist. Until recently he refused to use even autocomplete, preferring to type everything manually in vim - a text editor favored by hardcore programmers who navigate entirely through keyboard shortcuts. "I'm very pedantic about my code," he said.

What changed his mind was seeing the latest models in action. While he still would not hand over complex architectural decisions to AI, Pachocki said Codex has become indispensable for rapid experimentation. "I can have it run experiments in a weekend that previously would have taken me like a week to code," he said. "Once you see it do something that would take a week to do - I mean, that's hard to argue with."

The pattern mirrors what is happening at other leading labs. Anthropic has reported that its AI now writes the vast majority of its own training code, and similar tools have reshaped workflows across the industry. The real-world impact of such agent-based systems has already been measured: a Georgetown University study found that AI embedded in the Pentagon's Maven system enabled 20 people to do the work of 2,000.

Pachocki said the shift goes well beyond coding convenience. OpenAI now views Codex as a primitive version of the autonomous researcher it plans to build by 2028. "I don't think it is at the level where I would just let it take the reins and design the whole thing," he said. But the trajectory is clear: if most of OpenAI's technical staff already delegate substantial coding tasks to agents, the gap between a coding assistant and an autonomous researcher is shrinking fast.

Also read: From Chat Tool to War Machine: How 20,000 Troops Came to Depend on Claude AI.