What happened
On October 22, 2026, OpenAI announced that Uber is rolling out ChatGPT Work to its global workforce, more than 30,000 employees, starting that day. ChatGPT Work is an agent inside ChatGPT Enterprise that takes on multi-step tasks and returns finished work. OpenAI calls it Uber's first company-wide deployment of an agentic platform built for knowledge work. The rollout follows a pilot of 1,000 employees. It connects to Google Drive, Gmail, Google Calendar, and Slack, and employees can use it to research a question, prepare a meeting brief, build a presentation, or analyze data. GPT-6.1 Sol is the default model. Uber's engineers will keep using Codex for software work. OpenAI says the rollout is supported by an internal AI skills library, approved integrations, and employee training. Uber's VP of engineering, Sumanth Sukumar, told TechCrunch that about 80% of the pilot users were not engineers, and that they used the tool to analyze operations data, localize training material for different markets, and build personal dashboards. Two things were not announced. Uber did not disclose what the rollout costs, and neither company has published results. OpenAI says the companies plan to share insights on adoption, use cases, and measurable outcomes as the rollout progresses.
Why it matters for business owners
Most owners have the same question Uber had: the pilot looked good, so do we give it to everyone? Uber's answer is yes, but with structure around it. The useful part of the announcement is not the headcount. It is the list of what sits around the tool: approved integrations, a library of reusable skills, and training. Those are the parts a smaller business can copy without a large budget. It also shows who used the pilot. If the 80% non-engineer figure from Uber is accurate, the early value came from finance, operations, and product teams doing analysis and localization work, not only from developers. Finally, this is a vendor announcement made jointly with the vendor. It is a good description of a plan. It is not an independent measurement of whether the plan worked.
What owners should not misunderstand
Do not read "available to 30,000 people" as "30,000 people are getting value." Access is a number Uber can announce today. Outcomes are what Uber says it will study later. Do not assume your pilot results will carry over to the whole company. A pilot is usually staffed by people who volunteered and like the tool. Wider groups use it differently, and some will not use it at all. Do not copy the scale. Uber has an internal AI engineering capability. TechCrunch reports Sukumar described pairing 30 AI engineers with teams for two weeks earlier this year. A company with 80 people needs a smaller version of that support, not the same one. Do not connect every tool because the product allows it. An agent that reads email and chat can see whatever the signed-in user can see. Uber says its integrations are approved, which means someone decided which ones to turn on. And do not skip the cost question because Uber did. Uber has not said what this costs. Yours will not be a secret to your finance lead for long.
The operational lesson
Widening AI access is a rollout, not a purchase. The order matters more than the vendor. A pilot answers one question: does this help real people do a real job? A rollout answers a different one: can the whole company use it safely, consistently, and at a cost we accept? Many businesses treat a successful pilot as the answer to both. Uber's own wording points at the gap. It says it will look at how employees use the technology, what workflows they build, and how it helps them work more effectively, then use that to guide deployment. In other words, the learning continues after the rollout starts. That only works if someone owns the question and the measurement exists before access widens. The practical version for an owner: decide what "working" means before you add people. If you cannot say it in one sentence, you are buying access, not results.
What a serious business should do next
Start by writing down what your pilot actually showed. Which jobs, which people, how much time, how much rework. If the pilot was informal, run a short, defined one first. Choose the next group on purpose. Pick one or two teams with a clear repeating task rather than opening the tool to everyone at once. Approve the connections. List the mailboxes, drives, and chat channels the tool can reach, turn on only what the task needs, and ask your vendor in writing how permissions and data retention work for your plan. Train on real jobs. A short session on "the three tasks we want you to try, and how to check the output" beats a general AI talk. Capture what works so the next person does not start from zero. Name an owner for outcomes. One person decides what is measured, reviews it monthly, and can say stop. Track accepted work and rework, not only logins. Set a spending view before widening access. Know the seat price, any usage charges, and who approves more. Our post on Uber's earlier AI budget problem shows why. And leave a way back. Keep the old process available until the new one has proven itself on your own work.
The Atlacis view
Atlacis helps owners slow down, understand the workflow, map the risk, and choose the right tool or system before spending. Uber's announcement is a reminder that the hard part of AI is rarely the tool. It is deciding who uses it, on what, with which data, and how anyone will know it worked. If you have a pilot that went well and are deciding how far to take it, a short conversation can help you pick the next group, the connections to approve, and the measure that matters.
The short version
- On October 22, 2026, OpenAI and Uber announced ChatGPT Work for more than 30,000 Uber employees, up from a 1,000 person pilot.
- Uber says the rollout comes with approved integrations, an internal AI skills library, and training. It did not disclose cost and has not published results.
- Access is not adoption, and a pilot does not prove a rollout. Decide what "working" means before you add people.
- Widen in steps: choose teams on purpose, approve connections, train on real jobs, and name one owner for outcomes.
- Know the cost model and keep the old process available until the new one proves itself on your own work.