By Adri 
Generative AI adoption within large companies often begins with experiments.
One team uses an assistant to summarize documents. Another analyzes business data. Developers experiment with coding tools.
Scaling those experiments across an organization is more complicated.
Employees need training, systems must enforce access permissions, and management needs ways to evaluate results.
Uber is now taking that step with ChatGPT Work.
On October 22, the company and OpenAI announced the expansion of their partnership to cover more than 30,000 employees across Uber's global workforce.
The announcement follows an initial pilot involving 1,000 employees.
OpenAI describes the expansion as Uber's first company-wide deployment of an agentic platform specifically designed for knowledge work.
The announcement confirms the start of a broad rollout.
It does not establish that every employee is already using the service daily or that every internal business process has been automated.
A pilot involving 1,000 employees expands company-wide
Uber previously tested ChatGPT Work with approximately 1,000 employees.
Participants came from several functions, including operations, finance and product teams.
Sumanth Sukumar, Uber's vice president of engineering, told TechCrunch that approximately 80% of the pilot users were not engineers.
That is notable because early enterprise AI adoption has often been closely associated with software development.
Uber is extending agent capabilities into functions that manage the company's daily operations.
The wider rollout potentially covers employees working in different countries, departments and software environments.
The challenge is therefore not simply providing access to a new application.
It is finding tasks that are appropriate for automation across different kinds of work.
Uber has not defined a single mandatory workflow that every employee must follow.
Sukumar described an approach that gives teams access to powerful tools while allowing them to experiment with applications suited to their responsibilities.
The company is moving into a phase where practical use cases and results must be established at a much larger scale.
ChatGPT Work is designed to complete tasks rather than only answer questions
OpenAI introduced ChatGPT Work in July 2026 as an agent for extended, multi-step professional tasks.
Users can assign objectives that involve research, analysis and work across connected applications.
The difference between a chatbot and an agent is important.
A conversational assistant might explain how to build a presentation.
An agent with the necessary tools and permissions can gather information, organize the content and create the presentation itself.
Similarly, answering a question about a document is only one possible part of a more involved workflow.
A complex task might require analyzing several files, verifying figures and preparing a report for colleagues.
ChatGPT Work is intended to coordinate those activities.
Employees can monitor progress and provide clarification or approval when required.
The platform is therefore more than a conventional search interface connected to corporate documents.
It is designed to perform parts of the digital work traditionally carried out manually.
Google Drive, Gmail, Calendar and Slack provide working context
An enterprise agent's usefulness depends heavily on the systems it can access.
Company information is distributed across email, shared documents, calendars, messaging applications and specialized software.
ChatGPT Work supports connections to tools including Google Drive, Gmail, Google Calendar and Slack.
For example, an agent preparing a meeting brief could retrieve relevant documents and organize information available through approved integrations.
It can also help analyze data and prepare reports or presentations.
The announcement does not mean every Uber employee will automatically receive unrestricted access to all corporate information.
Uber and OpenAI refer to approved integrations and internal deployment arrangements.
Access permissions remain central to the security of this type of system.
An agent operating across multiple applications can combine information that employees previously accessed through separate workflows.
That capability can be useful, but it also makes permission boundaries especially important.
Organizations need to determine which data an employee may access and which actions an agent may perform on that employee's behalf.
Pilot users analyzed operations and built dashboards
Uber has provided several examples of how participants used ChatGPT Work during the pilot.
According to Sukumar, teams analyzed operational information to identify trends and develop recommendations.
Other employees used the system to localize training materials for different markets and languages.
Creating personalized interactive dashboards was another reported application.
These are familiar business tasks rather than entirely new categories of work.
The potential change is the ability to delegate parts of their execution to software capable of working with documents and data.
A dashboard may require combining different information sources, selecting useful metrics and preparing a usable interface.
An agent may reduce some of the manual effort involved.
However, the quality of the result still depends on the accuracy of the underlying information and the way it is interpreted.
Uber has not published a comprehensive pilot-wide success rate or independent evaluation of the errors generated.
The examples demonstrate reported applications, not the reliability of every workflow.
One capital allocation process fell from 15 hours to 30 minutes
Uber has highlighted another result from its internal AI Pods initiative.
The program brings AI engineers together with operating teams to identify opportunities for automation.
According to a public statement from Uber, one team reduced the time required for a capital allocation process covering 150 cities from fifteen hours to thirty minutes.
That represents a substantial improvement for the specific process.
It should not be generalized without additional evidence.
AI Pods is a broader automation initiative rather than a measurement limited exclusively to ChatGPT Work.
The public statement does not provide enough detail to attribute the entire improvement to the newly deployed agent.
It also does not fully establish how review and approval time was counted.
The example therefore does not prove that every Uber employee will become thirty times more productive.
It suggests that some workflows can be accelerated substantially when engineering specialists collaborate with business teams.
Whether similar gains can be achieved across other functions remains an open question.
GPT-6.1 Sol will be the default capability for Uber
The partnership also provides access to OpenAI's broader model portfolio.
According to the October 22 announcement, GPT-6.1 Sol will be deployed as the default capability within ChatGPT Work for Uber.
OpenAI presents the model as balancing advanced performance with cost efficiency.
Uber teams will also have access to relevant multimodal capabilities.
Voice and image generation are among the features mentioned.
Those capabilities can support different business activities, although their inclusion in the available technology does not mean every employee will necessarily have unrestricted access to every feature.
Actual use depends on the integrations and configurations approved by Uber.
The default model choice is still significant.
Across tens of thousands of employees, task frequency, input volumes and computing requirements can materially affect the cost of a deployment.
Balancing model capabilities and operating expenses becomes an organizational decision as much as a technical one.
Uber's software engineers will continue using Codex
The company-wide expansion of ChatGPT Work does not replace all existing AI tools at Uber.
OpenAI says the company's engineering and software development teams will continue using Codex.
The distinction reflects different requirements.
ChatGPT Work is designed for general professional tasks, from research and document analysis to finished deliverables.
Codex focuses on coding, code modifications and software engineering workflows.
Both can play roles within the same organization.
A product team might use ChatGPT Work to prepare an analysis while developers use Codex for implementation.
The announcement therefore represents a broadening of Uber's AI strategy rather than the replacement of every specialized tool with a single interface.
Internal AI skills and training will support adoption
Uber plans to support the rollout with an internal AI skills library.
The company also plans dedicated employee training and approved integrations.
These elements are important when moving from a general-purpose product to widespread organizational use.
A financial analyst, an operations manager and a marketing employee have different information needs.
Shared internal guidance can help employees identify relevant workflows.
It can also reduce the fragmentation created when departments independently develop incompatible processes.
Training has another purpose.
Employees need to understand how to define objectives, provide context and verify agent-generated results.
An agent may produce a polished report while relying on outdated information or an incorrect interpretation of a request.
Users therefore need to understand where automation is appropriate and where additional checks remain necessary.
Uber has not publicly disclosed every internal policy that will govern these activities.
The existence of training should not be treated as a guarantee that all operational or security risks have been eliminated.
Corporate data permissions remain a central challenge
Enterprise agents become more useful when they can access relevant organizational context.
That can include internal documents, operational information and previous work.
However, organizations must enforce appropriate access restrictions.
A global company such as Uber handles data relating to employees, commercial operations, markets and business partners.
Not all of that information should be accessible to every employee.
Agents connected to multiple applications must respect the boundaries between users and systems.
Those restrictions cannot depend solely on instructions telling an AI model to behave appropriately.
They also need to be enforced through application permissions, identity management and administrative controls.
Access to information must also be distinguished from permission to change it.
Reading a document and modifying a business record carry different consequences.
Sensitive workflows may require human confirmation.
OpenAI describes ChatGPT Work as a system in which users can monitor progress and approve important actions.
The practical scope of those controls will depend on Uber's configurations and internal policies.
Uber and OpenAI already collaborated on customer-facing products
The partnership did not begin with ChatGPT Work.
In May 2026, OpenAI described several AI-powered experiences developed with Uber for drivers and riders.
Uber was using OpenAI models to simplify certain interactions within its applications.
The company had developed assistants intended to help drivers understand earning opportunities and marketplace conditions.
It was also working on voice experiences that let customers express requests more naturally.
Those applications concern services provided to riders, drivers and other platform participants.
The October 22 deployment is different because it targets Uber's internal workforce and professional processes.
Uber is extending an established technology relationship into another part of its business.
The image illustrates an earlier OpenAI-powered voice feature in the Uber application. It is not a screenshot of the internal ChatGPT Work deployment.
Neither contract value nor company-wide savings have been disclosed
The announcement also contains important omissions.
Uber has not disclosed the financial value of the agreement.
The company has not provided a per-employee cost or an expected annual budget for ChatGPT Work.
The cost of the rollout therefore cannot be reliably calculated from the number of employees alone.
Uber also has not published a comprehensive assessment of productivity gains across the pilot.
There is no company-wide before-and-after measurement establishing how much employee working time has been saved.
Economic benefits will depend on frequency of use, task characteristics, reliability and the time needed to verify results.
An agent might accelerate document preparation while still requiring substantial human review.
It could also improve access to information without immediately reducing expenditure.
The number of employees receiving access is therefore not a sufficient measure of return on investment.
Uber plans to study usage patterns, the workflows employees develop and the resulting effectiveness.
The deployment follows a separate workforce restructuring
Uber's recent organizational changes provide relevant context.
On September 2, CEO Dara Khosrowshahi announced a restructuring involving a reduction of approximately 10% of the company's workforce.
He described the changes as an effort to simplify management structures and concentrate resources on major priorities.
The timing may prompt questions about how AI adoption could affect future jobs and responsibilities.
However, the proximity of the restructuring and the ChatGPT Work announcement does not establish that the new agent caused those job cuts.
The September decision and the October deployment are separate developments.
Uber did not announce a new workforce reduction directly linked to ChatGPT Work on October 22.
The company presents the tool as a way to help employees complete tasks and develop new ways of working.
The actual consequences for jobs, responsibilities and organizational structure will need to be observed over time.
The real test will be sustained adoption and measurable results
Expanding from 1,000 pilot users to more than 30,000 employees is a significant change in scale.
But several indicators will determine whether the deployment produces meaningful business results.
The first is regular usage rather than the number of people who merely have access.
The second is the frequency and complexity of tasks successfully completed by the agent.
The third is output quality, including the amount of human work required to correct mistakes.
Finally, any gains must be compared with the cost of the service, training and integration.
OpenAI and Uber say they intend to share further information about adoption, use cases and measurable outcomes.
They have not provided a specific timetable for those disclosures.
As of October 23, 2026, the confirmed development is the start of Uber's global ChatGPT Work rollout.
The company has moved beyond a limited pilot, but company-wide operational results have not yet been established.
For OpenAI, the agreement provides an important reference customer for enterprise agents.
For Uber, it represents an attempt to make AI a resource shared across the organization rather than a collection of isolated projects.
The next challenge is not distributing more access. It is demonstrating that agents can perform useful work reliably and at sustainable cost across very different professional roles.