Sam Altman Thanks "Original Devs" in Rare Public Tribute—Then Announces Their Replacements

The event was billed internally as a celebration. Sam Altman took the stage at OpenAI's San Francisco headquarters last Thursday evening in front of roughly 800 employees, a large screen behind him displaying a timeline of the company's technical milestones from 2018 to the present. He spent the first twelve minutes doing something he almost never does in public: naming names. He called out the individual engineers and researchers who built the foundational models—GPT-1, GPT-2, GPT-3—that transformed OpenAI from a research lab into the most consequential AI company on the planet. He thanked them for the "unreasonable bets" they made, for the late nights, for the code that "nobody thought would work until it did."

Then, in a transition that several attendees described as "jarring," Altman pivoted. He introduced what he called OpenAI's "AI-first development initiative"—a new organizational structure in which the company's core engineering teams will be rebuilt around AI-augmented workflows, with significantly smaller headcounts and a fundamentally different skill profile. The engineers who built the early models, many of whom were sitting in the front rows, were learning in real time that the company they helped create was moving on from the kind of work they do.

"It was surreal," said one attendee who spoke on condition of anonymity. "He literally thanked people by name for building the thing—and then explained that the thing they built is now good enough to replace how they built it. The room didn't know whether to clap or not."

OpenAI declined to comment for this story. But internal documents reviewed by this publication describe the AI-first initiative as a restructuring of the company's engineering division that will reduce traditional software engineering roles by approximately 30% over the next 18 months, while creating new positions focused on AI system design, prompt architecture, and model evaluation. The net headcount is expected to remain roughly flat, but the composition of the workforce will change dramatically.

The initiative reflects a bet that Altman has been making with increasing conviction: that AI models are now capable enough to handle a significant portion of the software engineering work that has traditionally required human developers. Internal experiments at OpenAI, according to two people familiar with the results, have shown that teams using the company's latest code-generation tools can ship features in roughly 40% less time with 60% fewer engineers. The quality, they said, is "comparable or better" because the AI catches classes of bugs that humans routinely miss.

The reaction inside OpenAI has been mixed. Several of the engineers honored during the tribute portion of the event have already begun exploring opportunities elsewhere, according to people with knowledge of the conversations. Others see the restructuring as an inevitable and even exciting evolution. "If you're an engineer who's afraid of AI eating your job, you probably shouldn't be working at an AI company," said one current employee. "This is what we signed up for."

The broader industry is watching closely. OpenAI is not the first company to restructure around AI-augmented development—Shopify, Klarna, and several mid-stage startups have made similar moves in the past year—but it is the most symbolically significant. When the company that builds the tools announces that those tools are good enough to replace its own workforce, it sends a signal that reverberates far beyond one company's org chart.

"This is the moment that a lot of engineering leaders have been dreading," said Dr. Kenji Watanabe, a professor of computer science at Stanford who studies the economics of software labor. "Not because the technology isn't ready—it is—but because nobody has figured out what the transition looks like for the millions of developers who were trained in a pre-AI paradigm. Altman just showed us one version of that transition, and it's not gentle."

The concept of "AI-first development" as practiced at OpenAI means more than giving engineers better autocomplete. It means redesigning the entire development process around the assumption that AI handles the first draft of most code, most tests, and most documentation. Human engineers become reviewers, architects, and system designers rather than line-by-line coders. The skill set shifts from implementation to judgment: knowing what to build, how to evaluate whether the AI built it correctly, and how to intervene when it didn't.

For the engineers who built GPT-1 through GPT-3, the irony is acute. They are, in a very literal sense, the authors of their own professional disruption. Several of them hold patents and have published landmark papers. Their work made Altman a household name and OpenAI a $300 billion company. And now that work is being used to argue that fewer people like them are needed.

Altman, to his credit, appeared to recognize the tension. Near the end of the event, he returned to the tribute theme. "The people in this room built something that will outlast all of us," he said, according to multiple attendees. "That's not a consolation. That's the highest compliment I know how to give." The applause, attendees said, was polite but restrained. Several people left before the reception.