Non-engineers made up about 80% of early adopters in Uber’s ChatGPT Work pilot, according to Sumanth Sukumar, the company’s vice president of engineering. Finance, operations and product teams were among the participants, putting everyday business workflows at the center of the experiment.
Uber is now expanding that 1,000-person pilot to its entire global workforce. Its October 22, 2026 announcement says ChatGPT Work will become available to more than 30,000 employees, supported by an internal AI skills library, approved integrations and dedicated training.
For businesses considering a similar deployment, that support plan offers more guidance than the employee count. Uber has described how it intends to help people use the technology, though it has yet to publish the spending or measured productivity results needed to judge the business case.
The Pilot Tested Work Outside Engineering
Uber’s reported pilot uses were specific: analyzing operational data to identify trends and develop recommendations, localizing training materials for different markets and languages, and building personalized interactive dashboards.
Those examples suggest that a workplace-agent pilot needn’t revolve around software development. Each also calls for a different evaluation. An operational analysis needs defensible conclusions; localized training needs accurate language and meaning; a dashboard needs reliable data and useful functionality. Grouping all three under “AI usage” would obscure whether employees received usable work.
The 80% figure describes the composition of Uber’s early adopters, not the share of all non-engineering employees using the tool or their success rate. It supports recruiting business functions into a pilot, but does not establish that non-engineers gained more than engineers.
Software engineering teams will continue using Codex, the company says. ChatGPT Work will sit alongside that developer tooling as part of a wider toolkit, without replacing every specialized application.
Uber Is Expanding Support Alongside Access
Uber describes ChatGPT Work as an agent operating within ChatGPT Enterprise, designed to handle complex, multi-step tasks. Its announcement lists connections with Google Drive, Gmail, Google Calendar and Slack, with potential uses including information research, meeting briefs, presentations, data analysis and workflows across applications.
These capabilities come from the company’s description; they have not been independently verified in Uber’s deployment. A connection to an application alone does not demonstrate that a complete workflow runs reliably.
The internal skills library, approved integrations and dedicated employee training provide a route for spreading useful practices beyond individual experimenters. Uber’s announcement does not detail the library’s contents or establish how consistently teams will use it.
Sukumar also described an earlier two-week “AI pods” exercise, in which Uber paired 30 AI engineers with teams to automate parts of their jobs. Technical specialists worked with employees who understood the underlying tasks, giving workflow owners help translating a familiar problem into something a system can execute.
That pairing is a concrete implementation example. Whether the automations succeeded remains unquantified.
A Rollout Checklist, With Clear Evidence Limits
Uber’s documented choices provide a practical starting point for other employers. The following checklist combines those choices with recommended evaluation steps; it should not be read as a claim that Uber has already implemented every control.
-
Recruit across functions. Include employees who own operational, financial, training and product tasks, and extend participation beyond technical enthusiasts. Uber’s pilot shows that non-engineers can form most of an early cohort. It does not establish which recruitment method produces the best results.
-
Choose tasks with inspectable outputs. Start with work that produces something a reviewer can assess, such as an analysis, localized document or dashboard. Define what makes that output acceptable before scaling. A completed artifact and a correct, useful artifact are different outcomes.
-
Prepare reusable skills and approved connections. Uber says it is providing both. Other employers should specify which data sources a workflow needs, what access is permitted and which practices employees can reuse. Application availability alone is not a sufficient deployment specification.
-
Assign training and implementation support. Dedicated training and Uber’s engineer-team pairing offer two different support mechanisms. A practical rollout should explain when employees can work independently, when specialists should help and who reviews consequential outputs. Those review arrangements are recommendations, not disclosed Uber procedures.
-
Set baselines before claiming gains. Record task time, output quality, correction effort, recurring usage and relevant costs. Compare equivalent work, including human review. Uber has expressed an intention to evaluate outcomes, but the cited accounts do not provide a measurement framework or published results.
The Business Case Still Needs Numbers
Uber did not disclose its spending on the rollout to TechCrunch. The available accounts also provide no adoption targets or measured productivity gains from the pilot.
Sukumar said Uber would examine how employees use the technology, the workflows they develop and whether it helps them work more effectively. That stated intention does not establish that an outcome-tracking program is in place.
Company-wide availability is a starting condition. Sustained usage, consistent deployment across departments and completed automation remain unproven. The reported pilot examples also cannot establish a return on investment without costs and comparable outcome measures.
