Uber Puts All 30,000 Employees on ChatGPT Work: What the Pilot Showed

Uber is putting every employee on ChatGPT Work. On October 22, 2026, the company told TechCrunch that its entire global workforce, about 30,000 people, will now use OpenAI's enterprise agent, up from a pilot of 1,000. About 80% of the pilot users were non-engineers, according to Uber's VP of engineering Sumanth Sukumar. What Uber did not disclose is just as telling: no spending figure, no productivity numbers, and no fixed plan for how people must use it.

This post lays out what was said, what it implies, and how to read a rollout like this if you are deciding whether to move a team from pilot to company-wide. For background on the product itself, see our ChatGPT Work vs Codex complete guide.

TL;DR

What is ChatGPT Work, and where does Uber fit?

TechCrunch describes ChatGPT Work as an enterprise-focused AI agent that OpenAI introduced in July 2026 and that corporate teams can deploy for various workflows. We have followed it since launch: thread orchestration on mobile, a cloud browser that works in signed-in websites, and writing-style personalization. Usage milestones were covered in our post on Codex and ChatGPT Work reaching 8 million users.

Uber has separately written about running coding agents at scale; we covered that in How Uber runs coding agents cost-effectively. The new announcement is different in kind: it is about knowledge workers across the company, not engineers building products.

What did the pilot show?

Per Sukumar, the 1,000-person pilot covered workers across Uber's departments, including finance, operations and product, and a majority, around 80%, were non-engineers. He listed three kinds of task:

1. Analyzing complex operational data to identify trends and develop recommendations.
2. Localizing training materials for different markets and languages.
3. Building personalized interactive dashboards to manage daily work.

Notice what is missing. There is no claim of hours saved, tickets closed, or cost avoided. The evidence offered for expansion is breadth: many kinds of people found many kinds of use. That is a legitimate reason to scale a pilot, but it is an adoption argument, not an ROI one.

The 80% non-engineer share is the number worth dwelling on. Enterprise AI adoption has leaned on developers because coding agents have clear success checks: tests pass or they do not. A pilot dominated by finance and operations staff suggests the agent format is reaching tasks without that kind of automatic verification. It also raises the question of who checks the work, a theme in our coverage of the Nielsen and monday.com survey on managers finding AI work harder to evaluate.

"Room to experiment": how Uber describes the rollout

Sukumar was explicit that Uber has not decided exactly how employees will use ChatGPT Work. "Our approach is to make powerful AI tools available across Uber while giving teams room to experiment," he told TechCrunch. As an example, he said earlier this year Uber paired 30 AI engineers with teams across the company and gave them two weeks to automate key parts of their jobs. Those "AI pods," he said, showed how much opportunity exists when technical expertise is combined with deep understanding of a job.

That is a recognizable pattern: embed a small group of builders with domain teams, let them prototype automations, then give everyone the platform. It mirrors what OpenAI itself describes in its forward-deployed engineering work, which we covered in Colin Jarvis on the deployment gap. The lesson for other companies is that tool access alone rarely changes workflows; the pods supplied the translation between a job and an agent.

What did Uber not say?

Uber says it will look at "how employees are using the technology, the kinds of workflows they develop and how it helps them work more effectively," then use those evaluations to guide how AI is deployed across the business. In other words, the measurement comes after the rollout. If you are copying this playbook, build your measurement plan before the expansion, not after.

On controls, an agent that can act on a company's systems raises real governance questions. Our piece on OpenAI's data agent and ChatGPT Work enterprise safety covers what OpenAI has said about guardrails, and any buyer should ask the same questions about permissions and audit trails.

Why does it matter that this is OpenAI, and not a rival?

TechCrunch frames the deal as part of OpenAI's push to scale its enterprise business as it competes with Anthropic, noting that OpenAI has swiftly gained a loyal following with business customers. That framing is the reporter's, and we have no seat counts from either company to test it. What we can say is that large-company deployments are the proof points both labs now fight over; we covered a different side of that contest in Meta and Microsoft cutting internal Claude use. A 30,000-person logo is worth more to OpenAI than the revenue alone, because it is evidence that agents can survive contact with a whole company.

It is also worth noting what a single-vendor rollout means for Uber. Standardizing on one agent simplifies training, support and security review. It also concentrates risk: pricing changes, outages (we covered a ChatGPT Work outage on August 31) and policy shifts hit everyone at once.

Pilot to rollout: how fast was this?

TechCrunch says the pilot came "earlier this year" and that ChatGPT Work was introduced in July, which leaves a window of only a few months between product launch and company-wide commitment. The exact pilot dates were not given, so we cannot say how long Uber watched the pilot before deciding. A short window is common when a vendor and customer work closely together, and Uber says OpenAI worked closely with it on the integration, but it also means the evidence base is thin. One quarter of usage by self-selected early adopters tends to overstate how the other 29,000 people will behave.

That selection effect deserves a plain warning. Pilot volunteers are usually the most curious people in the company. Whether the median finance analyst or operations manager keeps using an agent six months later is the real test, and it is exactly the number Uber has said it will measure but has not yet reported.

What should other teams take from this?

1. Pilot with non-engineers. Uber's pilot was 80% non-engineers, and that is where it found its broad use cases. If your pilot is all developers, you will learn only about coding.
2. Pair builders with teams. The AI pods model, 30 engineers and two weeks, is cheap and produces concrete automations to show everyone else.
3. Decide the metrics first. Uber has not shared outcomes. Pick two or three measurable ones (cycle time, error rate, hours on a named task) before you scale.
4. Budget for usage, not just seats. Agents consume more compute than chat. Ask the vendor how usage is metered and capped.
5. Plan verification. When most users are not engineers, someone must check outputs. Define who reviews high-stakes ones, such as financial analysis.
6. Keep an exit. Keep prompts, workflows and dashboards documented so a vendor change does not erase your gains.

What we will watch

Three signals will show whether this rollout holds up: whether Uber or OpenAI publish measured results, whether other large employers announce similar company-wide deployments, and whether OpenAI discloses seat counts for ChatGPT Work. For OpenAI's own numbers on European adoption, see our write-up of its EU agentic AI report. We will update this post if Uber shares outcomes.

Related reading

• ChatGPT Work vs Codex: the complete guide
• How Uber runs coding agents cost-effectively
• OpenAI's data agent and ChatGPT Work enterprise safety
• Codex and ChatGPT Work hit 8 million users
• Forward-deployed engineering and the deployment gap
• Managers find AI work harder to evaluate

Details are accurate as of October 22, 2026; figures come from Uber's statements to TechCrunch.