Uber’s 30,000-Employee ChatGPT Work Rollout Is a Test of Enterprise AI at Scale

Uber is taking OpenAI’s workplace agent out of the pilot phase and making it available to more than 30,000 employees worldwide. The October 22 announcement is a significant enterprise-AI deployment, but it is not yet proof that the tool has improved productivity: Uber has not disclosed the rollout’s cost or settled results.

What Uber announced

OpenAI’s October 22 announcement says Uber is beginning a company-wide rollout of ChatGPT Work inside ChatGPT Enterprise. The stated scope is more than 30,000 employees across Uber’s global workforce. TechCrunch described the population as the company’s entire workforce, or roughly 30,000 people. Those descriptions are close, but neither source presents an independently audited headcount, so the most precise wording is more than 30,000 people made eligible for access.

The move expands an initial pilot involving 1,000 employees. OpenAI calls it Uber’s first company-wide deployment of an agentic platform for knowledge work. In practical terms, ChatGPT Work is positioned as software that can handle multi-step assignments rather than simply answer an isolated question. OpenAI says the rollout starts immediately and will be supported by approved integrations, an internal AI-skills library and employee training.

Uber’s chief technology officer, Praveen Neppalli Naga, presented the deployment as a way to give employees tools to innovate and solve problems. OpenAI President and co-founder Greg Brockman framed it as access to frontier models and agents across the company. Those are company statements about ambition, not measurements of business impact.

What the pilot showed

The pilot provides a baseline for adoption, but not a controlled productivity experiment. According to TechCrunch’s report, the 1,000 participants came from departments including finance, operations and product. Sumanth Sukumar, Uber’s vice president of engineering, told the publication that about 80% of early adopters were non-engineers. That matters because the deployment is not being described only as a coding initiative for software teams.

Sukumar gave several examples of how pilot users worked with the agent. They analyzed operational data for trends and recommendations, localized training material for different markets and languages, and built interactive dashboards for day-to-day work. These are reported use cases relayed by an Uber executive. They establish what teams tried, not what the company has proved at scale.

Uber also told TechCrunch that it had paired 30 AI engineers with other teams for two-week “AI pods” to automate parts of particular jobs. That experiment suggests Uber is testing a combination of a general-purpose assistant and function-specific process redesign. It does not establish that the same tasks can be automated reliably for tens of thousands of employees, nor does it show whether any roles or staffing levels changed because of ChatGPT Work.

What employees can do

OpenAI says ChatGPT Work can connect to Google Drive, Gmail, Google Calendar and Slack. With those connections, employees can research internal information, prepare meeting briefs, create presentations, analyze data and complete workflows across applications. The service is designed to return finished work, rather than only a conversational response. OpenAI’s July product announcement describes the broader product as an agent that can work across apps and files, break down complex projects and remain on a task for hours when needed.

The stated design changes the adoption question. A conventional chatbot can be evaluated by answer quality in a single interaction. An agent that reads company data, moves between applications and produces a document or dashboard also has to be judged on permissions, source selection, handoffs, review and the risk of an incorrect action. Uber’s announcement says employees will have approved integrations and training, but it does not specify the full permission model, which tasks require human approval or how sensitive information will be handled in every workflow.

OpenAI says GPT-6.1 Sol will be the default capability within ChatGPT Work, with multimodal features such as voice and image generation available for different needs. Uber’s developers and software engineers will continue using Codex for coding workflows. That division indicates the rollout is intended to cover general knowledge work while preserving a separate tool for software development.

Why the scale matters

A company-wide license or access decision is different from a showcase pilot. Giving a tool to more than 30,000 employees creates the possibility of common workflows, shared internal skills and accumulated usage data across functions. It also exposes the limits that are easy to miss in a small trial: uneven adoption, duplicated work, unreliable outputs, security exceptions and the cost of supporting users who need help.

The choice of finance, operations and product as early pilot areas is notable because those teams work with recurring analysis and business processes. If ChatGPT Work becomes useful there, Uber could learn whether a general agent can shorten preparation work without weakening review. But the company has not said that it has achieved such savings. The announcement says OpenAI and Uber plan to share more information about adoption, use cases and measurable outcomes as the rollout progresses.

The event is therefore best understood as a live test of enterprise AI distribution. It demonstrates that a large consumer-platform company is willing to make an agent broadly available. It does not yet demonstrate that every employee will use it, that outputs will be accurate enough for consequential work or that the economics beat narrower automation projects.

What is known, and what is not

The confirmed facts are relatively clear: the announcement date is October 22, the access population is described as more than 30,000 employees globally, the starting point was a 1,000-person pilot, and the named integrations include Google Drive, Gmail, Google Calendar and Slack. TechCrunch independently confirms the company-wide expansion and adds the approximately 80% non-engineer figure, the department mix and examples of pilot work. PYMNTS separately reports the same scope and pilot baseline.

The missing figures are just as important. Uber did not disclose its spending for the rollout to TechCrunch. It has not published a settled productivity result, an adoption rate after the pilot, a time-saved figure or a benchmark against human-only workflows. OpenAI’s announcement promises future reporting rather than presenting outcome data. Claims that the tool will help employees get more done should therefore be attributed to the companies or treated as projections.

There is also no public evidence in these sources that the deployment is a headcount-reduction program. The rollout could coexist with other AI automation efforts, but it would be inaccurate to infer layoffs, savings or changed staffing from this announcement alone.

The operational questions ahead

Uber’s next challenge is measurement. A credible evaluation would need to separate access from active use and activity from value. Useful measures might include completion time for defined tasks, rework and error rates, source-traceability, employee adoption by function and the cost of model usage and support. None of those measures has been reported for this rollout.

Governance will matter at least as much as model capability. Connecting an agent to mail, calendars, files and messaging can make work faster, but it also expands the consequences of bad permissions or an unchecked instruction. Training and approved integrations are a starting point. The public record still does not say how Uber will audit access, retain logs, review high-impact outputs or handle data that should not be used by an automated system.

Frequently asked questions

What exactly is Uber rolling out?

Uber is making OpenAI’s ChatGPT Work available across its global employee base through ChatGPT Enterprise. OpenAI describes it as an agent for multi-step work that can use connected apps and files and produce finished materials. The rollout is broader than a developer API or a limited departmental chatbot.

How many Uber employees are included?

The companies describe the population as more than 30,000 employees worldwide. TechCrunch calls it the entire global workforce and uses “some 30,000.” Because no source provides a headcount audit, this should be treated as an approximate access figure rather than an exact payroll count.

How large was the pilot?

The initial pilot involved 1,000 employees. TechCrunch reports participation from finance, operations and product, and says about 80% of pilot adopters were non-engineers. That percentage was supplied by Uber’s Sumanth Sukumar and is not presented as an independently audited survey result.

Has Uber proved that ChatGPT Work improves productivity?

No. The sources describe examples of use and the company’s plan to track workflows and outcomes, but they do not provide a settled productivity benchmark, a cost figure or a verified time-saving result. The rollout is an adoption milestone and a test of the product, not a published proof of return on investment.

Which tools does ChatGPT Work connect to?

OpenAI names Google Drive, Gmail, Google Calendar and Slack for the Uber deployment. The companies also say Uber developers and software engineers will continue using Codex for software-development workflows. The public announcement does not enumerate every approved integration or permission rule.