Uber began rolling ChatGPT Work out to more than 30,000 employees on 22 October 2026, expanding a pilot that had put the agent in front of 1,000 people earlier in the year. Uber’s own announcement describes both stages. It does not describe what the rollout costs, how many of those employees will use it, or anything it has produced yet. In the same post, Uber says the two companies plan to share insights into adoption, use cases and measurable outcomes as the rollout progresses.

ChatGPT Work is the agent OpenAI announced in July 2026, and it is included with ChatGPT plans from Free through Enterprise. At Uber it runs inside the company’s ChatGPT Enterprise account, Uber’s plan and workspace. Work is the agent layer inside that plan, so the cost question is how credits and usage limits behave inside Uber’s tenancy, not the price of a new seat.

Several of the most quotable details, including the “AI pods” and the share of non-engineer adopters, come from one interview that Uber’s vice president of engineering gave to TechCrunch. Uber’s post carries the scale and the tooling.

Uber’s own post is specific about scale, tasks and tooling

The announcement states that the rollout started on 22 October and makes ChatGPT Work available to “more than 30,000 employees across Uber’s global workforce”, expanding from “an initial pilot of 1,000 employees”. Uber calls it the company’s “first company-wide deployment of an agentic platform tailored for knowledge work”. TechCrunch’s headline rounded that to “all 30,000 employees” and its body says “some 30,000”; Uber’s phrasing, “more than 30,000”, is the one to use.

The post names the office tools Work connects to inside Uber: Google Drive, Gmail, Google Calendar and Slack. Tasks include researching information, preparing meeting briefs, creating presentations, analysing data and completing workflows. GPT-6.1 Sol is the default model inside Work, and on Enterprise and Edu accounts OpenAI notes that it is off until an administrator enables it. Voice and image generation are enabled, and Uber’s developers keep using Codex. The rollout is supported by an internal AI skills library, approved integrations and employee training, and Uber names operations, finance, marketing and product as the functions involved.

Uber’s CTO, Praveen Neppalli Naga, frames the deployment around the company’s builder culture. OpenAI’s president and co-founder, Greg Brockman, calls reaching tens of thousands of employees a significant step. Both statements appear in the announcement, and neither comes with a number for spend, seats or impact.

The pilot reached 1,000 employees; the 80% non-engineer figure comes from one interview

Both Uber’s post and TechCrunch report a pilot of 1,000 employees. Uber’s post gives no date for it, TechCrunch says only “earlier this year”, and the earliest dated public source for the pilot is 22 October 2026.

The figure that has travelled furthest, that about 80% of the pilot’s early adopters were non-engineers, comes from TechCrunch’s interview with Sumanth Sukumar, Uber’s vice president of engineering. It does not appear in Uber’s own announcement, so it belongs in the record as an Uber executive’s account given to one outlet, not as a measured adoption rate.

The same interview is the only public source for “AI pods”: Sukumar describes pairing 30 AI engineers with teams across Uber for two weeks to automate key parts of their jobs. No Uber or OpenAI documentation of the programme was found. His examples of how Work gets used, analysing complex operational data, localising training materials for different markets and languages, and building personalised interactive dashboards, also come only from that interview.

The cost question is credits and usage limits inside Uber’s Enterprise workspaces

OpenAI’s pricing documentation lists Work and Codex as sharing “the same pricing, credits, and usage limits”, and puts the Business plan at $20 per user per month billed annually, or $25 monthly with at least two users. Enterprise and Edu are quoted through sales, and on flexible pricing they “have no fixed rate limits. Usage scales with credits.” Uber has not said which arrangement it is on.

In Enterprise workspaces, Work draws from a shared pool of credits. Administrators set per-user monthly limits, group defaults, overrides, a workspace overage limit and alerts, and OpenAI’s documentation notes that the alerts “notify; they don’t stop spending”. A cost-planning example on the same page, annualised from one week of usage by first-time adopters it labels Marketing, gives a median of roughly $53 per employee per year, an average of roughly $309 and a 90th percentile of roughly $1,154, while stating that these aren’t fixed per-user prices. None of those figures is Uber’s, and neither company has published Uber’s credit allocation, usage limits or monthly bill.

Uber’s plan to publish results later is the clearest sign that none exist yet

Uber’s post commits the two companies to sharing “further insights into adoption, use cases and measurable outcomes as the rollout progresses”. Sukumar told TechCrunch that Uber will “look at how employees are using the technology, the kinds of workflows they develop and how it helps them work more effectively” and evaluate those outcomes to guide future deployments. Like his other quotes, that one comes only from the TechCrunch interview.

Adoption counts and workflow stories are the easy part of a rollout like this to collect, and neither one is a productivity result. A team planning something similar can ask for the measurement design before the seats go live: which tasks are compared, against what baseline, and who decides whether the deployment has earned its credits. Uber has said that work will come later.

Our guide to measuring AI return on investment shows how to choose that baseline and count accepted work, review time, and total cost.

What remains unknown

  • Uber’s spend, seat count and any credit or usage figure specific to its tenancy. Neither company has published one. A third-party headline that put “usage limits” at the centre of cost control could not be checked: the article body was unreachable and no Uber-specific limit appears in either company’s material.
  • Any measurement of impact. The commitment to publish outcomes later is the strongest available evidence that none exists yet.
  • The pilot’s dates. Uber’s post gives none; TechCrunch says only “earlier this year”. No public source dated before 22 October 2026 was found for the 1,000-employee trial or the 80% figure.
  • OpenAI’s own account. Its customer page about Uber returned an HTTP 403 error on 24 October 2026. A secondary report of that page says it likewise omits seat counts, which could not be independently verified.
  • Employee, union, works-council or regulator comment. Searches of web and news results found none, and silence in the sources available is not proof that no such comment exists.
  • Any causal link to Uber’s September job cuts. The company cut roughly 3,300 corporate roles on 2 September 2026, about 10% of corporate staff, in a restructuring that was widely reported and not attributed to AI by Uber. Several aggregator posts note that the ChatGPT Work rollout followed weeks later. No source links the two beyond the timing.