Uber opens ChatGPT Work to more than 30,000 employees

Uber is making ChatGPT Work available to more than 30,000 employees, expanding a roughly 1,000-person pilot into a companywide deployment announced October 22. The move takes OpenAI’s workplace agent beyond a limited trial and creates a larger test of whether software that completes tasks across business applications can become part of everyday corporate operations.

According to OpenAI’s announcement, the rollout starts today within Uber’s ChatGPT Enterprise environment. Availability across the workforce does not mean that every employee is already using the product. The announcement concerns employees, rather than a new entitlement for the drivers and couriers who earn through Uber’s marketplace.

From answering questions to completing work

ChatGPT Work is designed to carry out multistep assignments using connected applications and company information. OpenAI describes research, briefs, presentations and data analysis among the intended uses, with connections to services including Google Drive, Gmail, Calendar and Slack. That makes integration with the organization’s existing workflows central to the deployment.

Uber is also developing an internal library of reusable skills and providing training, while its developers continue using Codex. These are different routes into the same organization: a coding assistant serves software development, whereas the wider workplace rollout reaches functions such as operations, finance and marketing. The companywide initiative is therefore broader than expanding an engineering tool.

The practical distinction is the output expected from an agent. A conventional question might produce an explanation of a spreadsheet. An agent assignment could involve locating the relevant files, analyzing their contents and assembling a finished presentation. That is an illustrative workflow, not a disclosed Uber production example, but it explains why connected tools change the scope of employee review.

Access is the first milestone; useful adoption comes next

The companies plan to share further findings about adoption and measurable outcomes. They have not established through this announcement how much labor the wider rollout will save, or whether any savings will exceed the deployment’s full cost. Uber did not disclose the rollout’s cost to TechCrunch.

The relevant business test is completed work that remains useful after review. If an agent produces a report quickly but leaves an employee checking every figure and rebuilding its conclusions, the time saved at the drafting stage can overstate the overall benefit. Our enterprise AI return-on-investment analysis explains why licensing and implementation costs also belong in that calculation.

A useful comparison would also match tasks of similar complexity and account for training time. Otherwise, changes in the work assigned could be mistaken for improvements caused by the software. Managers would need to distinguish frequent use from improved performance. A task might be opened often because it is valuable, or because employees are repeatedly correcting the output. Comparing completion time, error rates and rework against an earlier process would provide stronger evidence than counting prompts alone. These are evaluation criteria, rather than results Uber has reported.

A broader partnership, with different permission boundaries

OpenAI’s May customer case study described separate Uber applications for drivers and riders, including marketplace guidance and voice-based ride booking. Those product features provide context for the relationship but are not the same as today’s employee deployment. A consumer assistant responding inside a transportation app and a workplace agent accessing corporate files operate in different information environments.

OpenAI’s product documentation describes enterprise controls over integrations, access and sensitive actions. Their existence does not remove the need to configure them for a specific organization. As our guide to enterprise AI data security discusses, a useful deployment must limit which information and actions each user can reach, as well as review what the system produces.

For Uber, the next meaningful milestone will be evidence that staff keep using the system for repeatable work with acceptable accuracy and lower overall effort. For OpenAI, the rollout is a substantial distribution commitment inside an established global business. It is a customer expansion today; the productivity and financial effects remain questions to be measured.