Uber Puts All 30,000 Employees on ChatGPT Work After a Pilot Where 80 Percent Were Not Engineers

Uber will hand ChatGPT Work to its entire global workforce of about 30,000 people, moving an internal experiment from a 1,000-person pilot into company-wide access. Uber and OpenAI announced the expansion on Thursday. OpenAI described it as the first company-wide deployment of an agentic platform for knowledge work, a claim that is difficult to verify but easy to understand as a positioning statement: most enterprise agent rollouts so far have been departmental.

The pilot was deliberately spread beyond engineering. It covered finance, operations and product teams, and roughly 80 percent of the early adopters were non-engineers, according to Sumanth Sukumar, a vice president of engineering at Uber. Three uses dominated. Employees analyzed complex operations data to find trends and build recommendations, localized training materials for different markets and languages, and assembled interactive dashboards to run their own daily work.

The mechanics are mundane in a way that matters for adoption. ChatGPT Work connects to Google Drive, Gmail, Google Calendar and Slack, and runs on a portfolio of frontier models with GPT-6.1 Sol as the default. It arrived in July as an enterprise-focused agent rather than a chat window, which is the product category every large vendor is now pushing: software that takes a multi-step task, touches the tools where the work actually lives, and reports back.

Uber has also tried a more manual route to the same goal. Earlier this year it paired 30 AI engineers with teams across the company and gave them two weeks to automate parts of those jobs. Sukumar called the groups AI pods and said they demonstrated the value of combining technical expertise with a deep understanding of a specific job, a combination that generic tooling does not provide on its own.

What Uber will measure is the open question. The company says it will track which workflows employees actually build and how those workflows change the way teams operate, and that those results, rather than a pre-declared productivity target, will shape later deployments. It did not disclose spending on the rollout, which leaves adoption as the only observable indicator for outsiders.

Greg Brockman, president and co-founder of OpenAI, framed the deal as a step toward broader access to frontier capability. The commercial logic runs both ways: Uber gets a single negotiated surface across 30,000 seats, and OpenAI gets a reference account large enough to argue that its agent stack survives contact with a company whose operations span 70-plus countries, a dispatch marketplace and a delivery network.

There is a reason to treat the announcement cautiously. Seat counts are the easiest metric in enterprise software and the least predictive of value, and Uber is explicit that the pilot produced findings rather than a benchmark. The harder test starts now: whether 30,000 people who are not engineers can turn an agent into changed workflows, or whether usage settles into the pattern that has dogged every productivity suite since the spreadsheet, broad installation and narrow habit.