Up to $7,000 a Day per Engineer: OpenAI Shows How Its Employees Switched to Working with AI Agents
·
07.09.2026 17:57
·
2 min read
OpenAI has offered a rare look inside its own labs, publishing statistics on how its team uses autonomous AI agents. By mid-August, the median researcher at the company was burning through computing resources worth $600 per day at public API rates, while the top 10% of most active developers exceeded $7,000 a day.
Three ‘Workdays’ of Agents for Every Human Day
In a special report, Research acceleration: The view inside OpenAI, the company stresses that these figures do not represent actual financial payouts. They reflect the virtual cost of tokens at the standard rates paid by regular corporate clients of the platform. Still, the numbers vividly illustrate a tectonic shift in model development: back in July, the median compute spend per employee was a modest $162.
The explosive growth stems from a move to parallel computing. While before the start of summer, the total runtime of algorithms lagged behind human hours, by the end of August the ratio had flipped: now, for every eight-hour workday of an engineer, there are 3.1 equivalent ‘workdays’ of digital agents. Researchers no longer sit alone over code; they coordinate swarms of bots in the Codex environment.
Automation has virtually wiped out internal tech support: the number of questions in the channel for resolving infrastructure issues has more than halved, and some teams have even closed their regular consultation hours, handing server configuration over to language models.
Managers Instead of Coders, and Plans for 2028
However, full independence of AI from humans remains a myth. According to the company, over half of the multi-hour tasks successfully completed by agents over the past six months required at least one intervention by a human developer to correct course.
As Business Insider reports, the role of researcher at OpenAI has evolved: employees are no longer rank-and-file executors but have become foremen who assign technical tasks to digital subordinates, monitor checkpoints, and filter out hallucinations.
The startup’s leadership believes the ‘virtual intern’ phase has been successfully completed. The next milestone on the company’s roadmap is set for March 2028: by then, OpenAI plans to build a fully autonomous AI researcher capable of independently formulating hypotheses and running complex multi-day experiments without human oversight.
