Modeling an AI jobs transition

OpenAI dropped a framework this week that tries to answer the question everyone in Washington and Silicon Valley keeps dancing around: which jobs actually get eaten by AI, and which ones just change shape. The team ran the numbers on 921 occupations covering roughly 148 million American workers, sorting each into buckets based on how exposed the tasks are to current AI capabilities.

The categories aren't just "safe" and "doomed." Some roles face genuine automation risk — the tasks are narrow, repetitive, and AI already does them well enough. Others land in a reorganization bucket, where the job survives but the day-to-day work gets rearranged around AI tools, shifting what people spend their hours on. A third group actually grows, riding the demand AI creates for oversight, integration, and new services. And a chunk of occupations, according to the framework, see minimal disruption at all, because the work leans on things AI still can't touch — physical dexterity, in-person trust, judgment calls that don't reduce to a prompt.

What's notable is the scale of the ambiguity. This isn't a report that spits out a tidy number like "47% of jobs at risk," the kind of headline that's circulated since a famous 2013 Oxford study. Instead it's granular, occupation by occupation, task by task, acknowledging that a nurse and a paralegal and a warehouse picker are exposed to AI in completely different ways and on completely different timelines. That granularity is the actual point — it's a tool meant for policymakers and researchers to poke at, not a soundbite.

The framework also implicitly makes a bet: that the transition will be messy and uneven rather than a clean before-and-after. Some occupations will hollow out fast. Others will take years to reorganize as companies figure out how to actually fold AI into workflows without breaking them. OpenAI publishing this kind of analysis is itself a signal — the company that builds the models is now trying to shape the conversation about what happens to the people whose jobs those models touch.