OpenAI’s new research covering more than 900 occupations reached a result that runs against common fears around AI and jobs. Workers in the 18% of occupations classified as having high automation risk were not the first to be pushed out. Their unemployment rate has been rising more slowly than that of lower-risk occupations, according to the report.
That high-risk group includes data entry clerks, bookkeepers, and customer service workers, roles often cited in discussions about AI-driven replacement. OpenAI said people in these occupations are already using AI for roughly three times the task volume seen in other jobs. AI exposure is already heavy. Even so, labor outcomes have not deteriorated at the pace many expected.
Four-way job split shows high exposure does not equal immediate displacement
The report sorts occupations into four broad groups. 18% fall into the high automation-risk bucket, 46% are described as minimally affected, 24% are expected to shrink while still requiring human leadership, and 12% are classified as expansion roles, including software development.
OpenAI’s argument is that high automation risk does not automatically translate into immediate job loss. When AI makes a service faster and cheaper, total market demand for that service can expand enough to offset the labor savings. In that scenario, efficiency gains do not necessarily reduce headcount right away.
Lower costs can expand demand instead of shrinking payrolls
The report uses programming as an example. If the cost of writing code falls, demand for software and development work can surge, increasing the scale of software jobs rather than cutting them. That dynamic helps explain why software development appears inside the 12% expansion group.
This makes the labor impact of AI less linear than many headline claims suggest. A tool can reduce the labor needed for a single task, yet lower prices and faster output can pull in more customers and more work. In OpenAI’s framing, that demand growth is currently absorbing part of the pressure that automation would otherwise place on employment.
Current AI use is still below one-quarter of theoretical capability
The report also warns against reading too much safety into today’s numbers. OpenAI said workers in high-risk occupations are using less than one-quarter of AI’s theoretical capability. The implication is simple: the deeper stress test has not happened yet.
The source article also points to a DWF Ventures report published on April 17, which said AI accounts for 19% of DeFi trading volume but performs at less than one-fifth of human levels in complex trades. The pattern is similar. AI adoption is moving into real workflows at a fast pace, while actual replacement remains constrained by capability limits. If usage climbs much higher, or if those limits shift outward quickly, the balance described in OpenAI’s study could change.