Uber gives all 30,000 employees OpenAI's ChatGPT Work agent after a 1,000-person pilot

What Uber announced

Uber is giving its whole global workforce of more than 30,000 employees access to ChatGPT Work, OpenAI's agent for office tasks. OpenAI announced the rollout on 22 October 2026, saying it starts the same day and expands an earlier pilot that gave the tool to 1,000 Uber staff. OpenAI calls it Uber's first company-wide deployment of an agentic platform built for knowledge work, as opposed to a chat assistant that only answers questions.

The announcement is short on numbers beyond headcount. Neither company disclosed what Uber is paying, how many seats are active, or how the contract is structured. TechCrunch reports that Uber declined to say how much it is spending and that it will instead track how employees use the tool, the workflows they build, and whether it helps them work more effectively, before deciding how to deploy AI more widely. OpenAI says the two companies will share adoption figures, use cases and measurable outcomes as the rollout goes on.

Uber's chief technology officer, Praveen Neppalli Naga, framed the move as handing builders better tools. "As AI transforms how we work, we want to give everyone the tools to innovate, solve problems and help shape Uber's future," he said in OpenAI's post. OpenAI president Greg Brockman called it a big step toward expanding access to frontier AI across tens of thousands of employees.

What ChatGPT Work does

ChatGPT Work runs inside ChatGPT Enterprise. OpenAI introduced it in July as an agent that companies can point at multi-step jobs and that hands back finished work rather than a draft answer. According to OpenAI's post, it connects to the tools Uber already uses every day, including Gmail, Google Drive, Google Calendar and Slack. With those connections staff can research information, prepare meeting briefs, build presentations, analyse data and run workflows that cross several apps.

By default the agent will run on GPT-6.1 Sol, which OpenAI describes as balancing frontier performance with cost efficiency. The Next Web notes that OpenAI released Sol last month at about a fifth of the price of its top model. Uber's teams can also use OpenAI's voice and image generation features and pick other models from OpenAI's portfolio when a task needs them. Uber's software engineers are not switching tools: OpenAI says they will keep using Codex, its coding agent, for development work.

The rollout comes with scaffolding of Uber's own. OpenAI says it is backed by an internal AI skills library, a set of approved integrations and dedicated employee training, aimed at operations, finance, marketing and product teams. That mix matters: an agent that can read mail and edit shared documents is only as safe as the permissions and integrations a company allows, and Uber has chosen to define those centrally rather than leave each team to wire up its own connections.

What the pilot showed

The pilot that preceded the rollout was not an engineering experiment. Sumanth Sukumar, Uber's vice president of engineering, told TechCrunch that about 80% of the 1,000 early users were non-engineers drawn from finance, operations and product. They used the agent to analyse operational data and spot trends, to localise training materials for different markets and languages, and to build personalised interactive dashboards for their daily work.

Uber has also been testing a more hands-on model. Sukumar said that earlier this year the company paired 30 AI engineers with teams across Uber and gave each pairing two weeks to automate key parts of a job. He said those "AI pods" showed how much opportunity there is when technical expertise is combined with deep knowledge of a particular role. Uber says it has not settled on exactly how employees will use ChatGPT Work and wants to give teams room to experiment. That is a different approach from the one Anthropic recently recommended to enterprises, which centres on carefully designed single-agent loops rather than broad access.

The cost question

Uber has been public about wanting more for its AI money. The Next Web reports that in August Neppalli Naga said the era of "tokenmaxxing", meaning spending freely on AI usage, was ending, and that Uber had reportedly used up its entire 2026 budget for Anthropic's Claude Code by April. Those reports are not confirmed in OpenAI's announcement, which does not mention Anthropic or budgets.

The choice of a cheaper default model fits that story. Running a company-wide agent on a lower-cost model, while keeping the option to call stronger models when needed, is one way to cap the bill as usage spreads beyond engineers. It also lines up with recent reporting that OpenAI is winning business customers from Anthropic with deep discounts. Neither company has said whether Uber's deal involved a discount, so the link is context rather than a confirmed term of the contract.

Jobs and the layoffs backdrop

The rollout lands weeks after Uber cut staff. Business Insider reports that last month Uber laid off about 10% of its corporate workforce, roughly 3,300 people, which chief executive Dara Khosrowshahi said was meant to reduce management layers and the number of small teams. That announcement did not cite AI. Some laid-off employees told Business Insider that AI was already doing significant parts of their jobs, from answering questions in Slack to drafting replies to riders and drivers.

An earlier round was explicitly tied to AI. In July, Uber cut about 10% of its customer-service team, the first time it linked layoffs to a push to use AI, according to The Next Web and Business Insider. Neither Uber nor OpenAI connected the ChatGPT Work rollout to job cuts, and the announcement presents it as a productivity tool for the people who remain. Still, the timing means employees and investors are likely to read the deployment against that backdrop, and the outcome data Uber has promised will be watched for signs of whether the agent replaces work or adds to it.

Why it matters for OpenAI

For OpenAI, Uber is a large, recognisable reference customer for its agent products at a time when it is pushing hard into enterprise sales. TechCrunch describes the company as once known mainly as a consumer brand that has quickly built a loyal business following. OpenAI's own report on European business usage said agent tools such as Codex and ChatGPT Work produced more than half of the output tokens at its EU business customers in September, so wide internal rollouts like Uber's are exactly the usage it wants to grow.

The deal also shows where workplace agents are heading. The pitch is no longer a chatbot window in a browser tab but an agent that reads a company's mail, calendar, files and chat, and then produces slides, briefs and analyses on its own. That makes integration, permissions and training as important as model quality, which is why Uber is pairing the rollout with an approved integration list and a skills library instead of simply switching it on.

What is still unknown

Several basic facts are missing. Uber has not said what it pays, whether every employee will get the same access or connectors, how data from Gmail, Drive and Slack is governed inside the agent, or which results from the 1,000-person pilot persuaded it to expand. OpenAI's quotes describe ambition rather than measured gains, and the only pilot details come from Sukumar's comments to TechCrunch.

The promised follow-up is the part to watch. If Uber and OpenAI publish adoption rates, time saved or error rates from the full rollout, it would be one of the few large public datasets on how an office agent performs across a whole company, including the 80% of staff who are not engineers. Until then, the rollout is a statement of intent: a 30,000-person bet that a cheaper default model plus company-set guardrails can make an agent useful for everyday work.