An agent for work beyond engineering
TechCrunch reported that approximately 80% of early users were non-engineers, citing Uber vice president of engineering Sumanth Sukumar. That makes the expansion relevant to teams considering agents for ordinary office work. A deployment can reach far beyond software development while still requiring substantial work to turn individual experiments into dependable processes. OpenAI says the Uber deployment will support connections to Google Drive, Gmail, Google Calendar and Slack. Intended tasks include research, meeting preparation, presentations and data analysis. Uber will support adoption with employee training, approved integrations and an internal AI skills library. Its developers will continue using Codex.What ChatGPT Work is designed to do
OpenAI introduced ChatGPT Work on July 9, 2026. This is therefore a customer expansion of an existing product. In its product announcement, OpenAI describes an agent that can gather information from connected applications, break larger assignments into steps and produce documents, spreadsheets, presentations or websites. Users can inspect progress, redirect work and approve important actions. These are advertised capabilities, rather than results independently measured inside Uber. Consider a hypothetical operations review: an employee asks an agent to assemble source material, investigate a change in performance and prepare a presentation. The useful outcome would be a reviewable deliverable whose figures and conclusions can be traced to the underlying evidence. The difficulty is evaluating the whole process. A polished presentation offers limited value if someone must spend longer checking its calculations than they would have spent preparing it manually.Access is the starting point for measuring adoption
The rollout announcement does not provide a measured productivity improvement, error rate or deployment cost. OpenAI says the companies intend to share adoption information and measurable outcomes as the expansion progresses. Those disclosures would help distinguish experimentation from sustained operational use. For example, a useful assessment would track whether employees return to the same workflow, whether the output passes review and how often a person must intervene. It should also include the time required to correct mistakes. Measuring prompts or accounts alone would leave those questions unanswered. A frequently used agent can still create expensive review work, while a narrowly used agent could deliver substantial value on a recurring task.Permissions and spending shape the outcome
OpenAI’s product documentation describes administrative controls over connected tools, company context and permitted actions. It also describes compliance visibility and spending controls for enterprise deployments. Their availability does not establish Uber’s specific configuration. For buyers, this creates a practical evaluation question: can employees complete useful work within appropriate access boundaries and a predictable budget? AgentsCalling’s analysis is that the strongest evidence would come from comparing repeatable tasks before and after deployment. Relevant measures include total completion time, factual corrections, failed tool actions, human approvals and cost per accepted deliverable. Uber’s expansion creates an opportunity to study workplace agents at scale. Evidence of lasting business value will come from what employees repeatedly accomplish with that access. Reporting note: This article compares OpenAI’s primary announcements with accessible independent coverage. TechCrunch’s indexed excerpts were accessible, but direct full-page retrieval was unavailable during verification. AgentsCalling has not tested Uber’s deployment or independently audited its results.About Puneet Sharma
Puneet Sharma · Freelance Writer
Puneet Sharma is a freelance writer covering AI agents, automation and the software changing how we work. His coverage at AgentsCalling starts with the questions a reader might ask over a conversation: What does this tool actually do? Where does it fall short? And is it useful beyond the demo? His articles explore new launches, practical workflows and the trade-offs behind giving software more responsibility, with explanations that make room for readers who are still getting familiar with AI.
