Uber is expanding ChatGPT Work across its workforce. TechCrunch reported the October 22 rollout.
That is a significant change in access, but access is not the same as adoption, and adoption is not the same as improved performance. The announcement moves the question from whether a limited pilot can attract users to how a large organization will evaluate the work those users actually complete.
From a pilot to company-wide availability
OpenAI’s original October 22 announcement says the expansion makes ChatGPT Work available to more than 30,000 Uber employees globally, following an initial pilot involving 1,000 employees. The vendor describes support from Uber’s internal AI skills library, approved integrations and employee training across functions including operations, finance, marketing and product.
The statement also describes connections to workplace services such as Google Drive, Gmail, Google Calendar and Slack, with intended tasks including research, meeting preparation, presentations and analysis. These are announced capabilities and deployment arrangements. They do not establish that every employee is already a regular user or that every suggested workflow has delivered a positive result.
The companies say they plan to share further information about adoption, use cases and measurable outcomes as the rollout progresses. That leaves an identifiable next milestone: evidence about use and results, rather than another increase in the number of people entitled to try the system. Regular use would still need a definition that distinguishes repeat work from occasional experimentation.
Earlier workplace research offers a narrower comparison
There is relevant research on workplace assistance, but it should not be turned into a forecast for Uber. The original public research abstract hosted by Stanford, dated April 2023, describes the staggered introduction of a conversational assistant among 5,179 customer-support agents. The researchers measured issues resolved per hour and reported an average productivity increase of 14%, with larger gains for less experienced or lower-skilled workers and little effect among the most experienced or highly skilled.
That study concerns a particular support setting and a defined performance measure. It is not an evaluation of ChatGPT Work, Uber’s departments or the present rollout. Its value here is methodological: it shows why a concrete task, a comparison and variation between workers matter when discussing productivity.
An organization could have enthusiastic users without the same measurable improvement in every team. Equally, a useful gain in one workflow might disappear in a company-wide average. Evaluating a deployment only through a single participation figure would make both patterns hard to distinguish.
The meaningful unit is completed work
For this rollout, an informative evaluation would connect usage to a finished task and its quality requirements. A faster first draft is not necessarily a faster approved deliverable if substantial checking or correction follows. Conversely, a task might be valuable because it becomes feasible at all, even when the time saving is small.
The same distinction applies to training. Providing instructions is one deployment activity; establishing that workers understand when to verify an answer or request review is a different result. The announcement establishes the presence of a training program, not the effectiveness of every part of it.
Comparisons would be strongest when they preserve the task’s acceptance criteria and include review effort. Otherwise, apparent speed gains could simply reflect accepting less complete work. Differences between roles should also remain visible rather than being flattened into a universal claim about what an employee can now do.
Uber’s expansion is therefore evidence of a broad organizational commitment to experiment with agent-assisted work. The commercial announcement does not yet supply a controlled estimate of the return. The most useful follow-up will show where access becomes sustained use, where sustained use becomes dependable output, and which workflows still need a different approach.
