Uber has handed a company-wide AI agent to roughly 30,000 people and, by its own account, has not yet decided what they should do with it. That is either reckless or refreshingly honest, depending on how much you trust a pilot program. On Thursday, Uber and OpenAI announced that ChatGPT Work is going out to the whole global workforce.
What was announced
ChatGPT Work is an enterprise-focused AI agent that OpenAI introduced in July. Companies can deploy it across a range of workflows. OpenAI's own summary says Uber's teams will use it to research, analyze data, build presentations and complete complex, multi-step tasks.
The headline number is the interesting part. Earlier this year Uber ran a pilot with 1,000 employees. Now the figure is the entire company, about 30 times larger. Uber says OpenAI worked closely with it on the integration, which is the sort of sentence that usually means a lot of engineers and a lot of meetings.
The pilot was not an engineers' club
The pilot group came from across the business, including finance, operations and product. According to Sumanth Sukumar, Uber's VP of engineering, about 80% of those early users were not engineers.
That matters because enterprise AI pitches often assume the first customers are developers. Here the early adopters were mostly people whose jobs involve spreadsheets, markets and planning. Sukumar listed what they actually did with the tool:
- Operations analysis. Digging through complex operational data to spot trends and draft recommendations.
- Localization. Adapting training materials for different markets and languages.
- Personal dashboards. Building interactive dashboards tailored to managing their own daily work.
None of this is glamorous. All of it is the kind of work that eats a Tuesday.
A rollout without a script
Uber is going all in on the tool, yet it has not fixed a single prescribed way to use it. Sukumar described the approach as making powerful AI tools available across the company while leaving teams room to experiment.
He offered one example of that philosophy. Earlier this year Uber paired 30 AI engineers with teams around the company and gave them two weeks to automate key parts of those teams' jobs. Uber calls these groups "AI pods." His takeaway was that the payoff is large when technical skill meets deep knowledge of a particular job.
It is a sensible instinct. Nobody at headquarters knows what a regional operations lead spends her week on, so asking her (with an engineer sitting next to her) beats issuing a memo. It also means that "30,000 employees now use ChatGPT Work" describes access, not outcomes.
What Uber is not saying
Uber did not disclose how much it is spending on the rollout. For a deployment at this scale, that is the one number everyone would like to see, and it is absent.
What Uber did promise is measurement. Sukumar said the company will watch how employees use the tool, what kinds of workflows they build and whether it helps them work more effectively. Uber will then evaluate those results and let them guide how it deploys AI across the business.
That is a reasonable plan, with one caveat. "We'll see what happens and learn from it" is a perfectly good way to run an experiment. It is a slightly unusual way to describe a purchase covering an entire workforce.
The competitive backdrop
The deal fits a larger story about OpenAI. The company was long seen mainly as a consumer brand, but it has been moving fast to build its enterprise business, partly to compete with Anthropic, its chief rival. Business customers have become a loyal part of its base, while other AI companies have struggled to keep pace.
A flagship customer deploying the product to everyone is the kind of reference that sells to other companies. Uber gets a tool for its staff, and OpenAI gets a story to tell the next procurement department.
Where the sources differ
One small discrepancy is worth flagging. OpenAI's summary says the rollout covers more than 30,000 employees. The other account says the entire global workforce, which it puts at some 30,000 people. The gap is probably rounding, but the exact headcount was not stated in either.
What to watch
The pilot suggests the early value sits with non-engineers doing analysis, translation and dashboard work. The open questions are what the rollout costs and whether the usage Uber plans to track shows real changes in how work gets done. For now, Uber has bought a very large set of seats and is promising to find out later what they are for. Plenty of software purchases have been made on worse terms.