Uber hands ChatGPT Work to 30,000 employees before showing what it delivers

Uber is rolling out the ChatGPT Work agent to more than 30,000 employees worldwide, after a pilot with 1,000 people. The announcement shows where workplace AI is heading: out of engineering and into finance, marketing and operations. What it does not show yet is any measured result.

A 1,000-person pilot becomes a company-wide tool

OpenAI said on October 22 that Uber is launching ChatGPT Work across its global workforce. It is an agent inside ChatGPT Enterprise built to take on multi-step tasks and hand back finished work. More than 30,000 people get access, up from a pilot of 1,000. According to OpenAI, this is Uber's first company-wide deployment of an agentic platform for knowledge work.

The agent connects to Google Drive, Gmail, Google Calendar and Slack. Staff are meant to use it to research, prepare meeting briefs, build presentations, analyze data and finish workflows that span several apps. The default model will be GPT-6.1 Sol, which OpenAI pitches as a balance between frontier performance and cost. Voice and image generation are included.

The rollout is backed by an internal library of AI skills, approved integrations and training for operations, finance, marketing and product. Developers keep using Codex. The announcement contains no adoption or impact figures. Both companies say they will share those as the rollout progresses.

Uber already knows what agents cost, so it picks one channel

Where Uber is coming from matters more than the seat count. In June, according to Bloomberg, it capped token spending on agentic coding tools such as Cursor and Claude Code at $1,500 per employee per tool each month. Simon Willison noted at the time that Uber had burned through its 2026 AI budget in four months.

Choosing a default model explicitly framed around cost fits that story. A company that got burned by open-ended usage is now building AI for everyone on one vendor, approved integrations and its own library of playbooks. For IT and procurement, that is far easier to govern than dozens of subscriptions teams buy on their own.

For OpenAI, Uber is another large reference customer for ChatGPT Work, which launched on July 9. For anyone selling agents into offices, it signals that the fight is moving from the IDE into spreadsheets, inboxes and calendars.

An agent inside your inbox is mostly a permissions problem

Connecting to Gmail, Drive and Slack means the agent reads and writes where contracts, internal debates and personal data about riders and drivers live. The announcement mentions approved integrations but says nothing about how Uber handles permissions, audit trails or what the agent may send without a human sign-off. Across 30,000 users, that is a large surface for prompt injection and accidentally shared documents.

There is also context OpenAI leaves out. On September 2, Uber announced about 3,300 layoffs, roughly 10% of staff, to trim management layers. Uber does not claim ChatGPT Work replaced anyone. But employees are receiving the tool while the company is visibly cutting, and that will shape how willingly they hand it their own work.

Hard numbers will decide this, not another seat count

The first signal is what Uber actually publishes. Hours saved, the share of active users after three months or cost per employee would be data. Another quote from a satisfied manager would just be marketing.

The second signal is budget. If Uber eventually caps ChatGPT Work spending too, it will mean agentic office work burns tokens much like coding does. The third is whether the agent takes hold in finance and operations, or stays with the people who already tried it in the pilot.

Lilith's verdict

In June, Uber watched its engineers' token tab like a bartender counting every pint after midnight. Now it is opening the bar to 30,000 people and promising to post the price list some time later.