Uber ChatGPT Work rollout: what 30,000 employees get

By Sidra Salman · October 23, 2026

On Thursday, Uber gave ChatGPT Work — OpenAI's enterprise AI agent — to its entire global workforce of roughly 30,000 people. Not a pilot group. Not one department. Everybody.

I cover AI tools for a living, and this one is different. A 30x jump — from a 1,000-person pilot earlier this year to the full company — tells you where enterprise AI is actually heading. Here's what happened, what Work mode does in a corporate setting, and what it means for the rest of us.

What ChatGPT Work actually does inside a company

Quick context, since I covered the consumer side on October 17: ChatGPT's new Chrome extension has a "Work" mode that acts on your browser. The Uber deal is the enterprise twin of that idea.

ChatGPT Work is an agent built inside ChatGPT Enterprise, designed for jobs too complicated for a normal chat prompt. According to OpenAI's announcement, it can:

• Act across your apps and files — at Uber, that means Gmail, Google Drive, Google Calendar, and Slack.
• Work multi-step tasks for hours if needed — not just fire back a paragraph.
• Deliver finished work — sheets, slides, documents, web apps, meeting briefs, presentations, data analysis.

What Uber did, step by step

The pilot. Earlier this year, 1,000 employees from finance, operations, and product tested ChatGPT Work — and roughly 80% of them were non-engineers, Uber VP of engineering Sumanth Sukumar told TechCrunch. They used it to analyze operations data, localize training materials across markets and languages, and build personal interactive dashboards. Before scaling up, Uber paired 30 AI engineers with teams for two-week "AI pods" to automate key parts of people's jobs. The company has no fixed playbook: "Our approach is to make powerful AI tools available across Uber while giving teams room to experiment," Sukumar said. Uber will let the winning workflows guide its future AI spending.

The setup. GPT-6.1 Sol — the model we broke down from OpenAI's DevDay announcement — is the default model, which OpenAI says balances frontier performance with cost efficiency. Engineers keep using Codex for coding. The rollout includes an internal AI skills library, approved integrations, and training. Uber did not disclose the cost.

How big is this, really? The biggest enterprise AI rollouts

Here's the leaderboard, as reported by the companies themselves:

Company	AI tool	Headcount	When
Uber	ChatGPT Work (OpenAI)	~30,000	October 2026
PwC	ChatGPT Enterprise (OpenAI)	100,000 (US, UK, Middle East)	May 2024
Accenture	Microsoft 365 Copilot	~743,000	April 2026

A few things worth noticing. Accenture's Copilot deployment is the largest AI rollout ever — but Copilot is mainly an assistant inside Office apps, and only about 3% of Microsoft's 450 million-plus enterprise users pay for it, per Reuters. PwC's 100,000-seat ChatGPT Enterprise deal (May 2024) was OpenAI's biggest customer win at the time. Uber's is smaller in headcount, but it's one of the largest single-company deployments of an agent — software that acts across your apps — rather than a chatbot you talk to. That's the shift that matters.

The trade-off nobody's hyping

Here's the part worth a second read: ChatGPT Work connects to Gmail, Google Drive, Google Calendar, and Slack — emails, documents, calendars, and internal conversations flowing through one AI agent. Enterprise controls and training help, but before signing, I'd want these answered:

• What exactly can the agent read? "Connected to Gmail" could mean your inbox, sent folder, shared mailboxes, or everything. Permissions need to be specific.
• Where does the data go? Enterprise tiers typically exclude your data from model training, but get it in writing.
• Who can see what the agent produces? An agent that summarizes a Slack channel for your manager is a different product than one that works only for you.
• What happens when it gets it wrong? Agents that act — sending emails, editing files, updating dashboards — can make confident mistakes at scale. We covered what happens when AI agents go rogue earlier this month: the more an agent can do, the more carefully you need human review of what it did.

Should your company do this? A 5-question framework

Uber's rollout is a signal, not a template. Before copying it, run through these five checks:

1. Do you have repeatable workflows, or just chat curiosity? Uber's pilot found three concrete uses: data analysis, localization, and dashboards. If your team mostly asks the AI questions, an agent rollout is premature. You need jobs, not vibes.

2. Is your data house in order? An agent is only as good as what it can reach. Connect it to a clean, well-permissioned slice of your data first.

3. Can you run a real pilot? Uber's 1,000-person pilot across departments — plus two-week "AI pods" pairing engineers with business teams — is the model. Measure what people actually build, then scale what works.

4. Can leadership tolerate "no fixed playbook"? Uber let teams experiment and learns from the workflows that emerge. A 40-page usage policy before anyone clicks anything will strangle the experiment.

5. Have you answered the security questions? See the section above. If you can't answer all four, you're not ready for company-wide.

What this means for you, the worker

Forget the corporate strategy. The career signal is clear — and it's not "learn to code." 80% of Uber's pilot users were non-engineers: finance analysts, operations staff, product managers — people who can describe a job clearly and check the output.

Here's what I'd actually do this week:

• Learn to brief an agent, not just prompt a chatbot. Write the task the way you'd brief a junior colleague: goal, context, constraints, what "done" looks like.
• Pick one recurring workflow and automate it. A weekly report, a data cleanup, a meeting-brief template. Measure the time saved — that's your portfolio.
• Get comfortable with data. The most common enterprise-agent job is still "look at this data and tell me what matters." Our AI data analysis tools comparison is a decent starting point.
• Always verify. Agents produce finished-looking work that can be confidently wrong. The person who catches the mistake is more valuable than the person who generated the draft.

One more piece of context, reported by Tech Times: Uber cut roughly 3,300 jobs in September 2026, which CEO Dara Khosrowshahi attributed to cutting management layers rather than AI. The company hasn't said ChatGPT Work will replace roles — but the sequence is worth noting: restructure first, give everyone an AI agent second. The workers who learn to direct these agents are the ones this story favors.

Frequently Asked Questions

Is ChatGPT Work the same as regular ChatGPT?

No. Regular ChatGPT answers your questions. ChatGPT Work is an agent inside ChatGPT Enterprise that takes on multi-step tasks, acts across connected apps and files, and delivers finished work like slides, sheets, and documents. OpenAI began rolling it out in July 2026.

Why did Uber pick GPT-6.1 Sol?

Per OpenAI's announcement, it balances frontier-level performance with cost efficiency — a sensible pick when every query across 30,000 users has a price tag. Uber's engineers keep using Codex for coding instead.

Is Uber's rollout really the biggest of its kind?

It's among the largest single-company deployments of an AI agent, but not the largest AI rollout overall — Accenture's Microsoft 365 Copilot deployment covers around 743,000 employees. The meaningful distinction is assistant versus agent: Uber's people get software that acts across their apps.

About the Author

Sidra Salman is the founder of AI Tools Hub, where she writes practical, research-backed guides on AI tools and how to use them. Every guide is tested and written in plain English for a US audience.