Uber announced together with OpenAI on October 22, 2026, that it is rolling out ChatGPT Work, an enterprise agent platform, to its entire global workforce of more than 30,000 employees. The tool had previously been piloted with just 1,000 people inside the company, so this expansion is roughly a thirtyfold jump in seats.
Most of the people doing real work with it were not engineers
ChatGPT Work, launched by OpenAI in July, runs inside ChatGPT Enterprise and connects to the apps employees already use, including Drive, Gmail, Calendar and Slack. It can research topics, prepare meeting briefs, build presentations, analyze data and carry multi-step tasks across applications to completion. Sumanth Sukumar, Uber's vice president of engineering, said about 80% of the pilot users were not engineers, coming instead from finance, operations, product and other teams.
Three uses kept recurring in the pilot: analyzing operational data to spot trends and develop recommendations; adapting training materials for different markets and languages; and building personalized interactive dashboards to manage daily work. Uber CTO Praveen Neppalli Naga framed the expansion simply: as AI changes how work gets done, the company's job is to put the tools in everyone's hands so people can innovate and solve problems themselves.
Handing out tools is easy; the accounting comes later
Notably, Uber has not prescribed one company-wide way to use the tool, leaving teams room to experiment. Earlier this year it also tried AI pods, pairing 30 AI engineers with teams for two weeks to automate parts of their jobs and surface opportunities from real roles. The cost of the wider rollout was not disclosed. Uber says it will track how employees actually use the platform, which workflows they build and whether it genuinely helps, before deciding what to fund next.
For other companies weighing a workforce-wide AI rollout, the useful lesson is the order of operations rather than the headline number: validate with 1,000 people that non-technical staff can find real uses, then scale to everyone, and judge success by actual usage instead of accounts issued.
