The takeaway
The interesting part of Uber's rollout is the plumbing, not the seat count: an internal AI skills library, an approved integration list and two-week AI pods pairing engineers with business teams. That is the layer most agent programmes skip.
Why it matters for builders
If you are rolling agents into a company, copy the boring parts: a curated skills catalogue, a whitelisted integration set, training, and a small pod of engineers embedded with business teams. Model choice (here GPT-6.1 Sol by default) is the cheap decision; permissions and connectors decide whether the deployment actually ships.
Uber rolls ChatGPT Work out to all 30,000 employees
Uber has moved its OpenAI pilot from a 1,000-person experiment to a company-wide deployment. On Thursday the ride-hailing giant confirmed that every one of its roughly 30,000 employees now has access to ChatGPT Work, OpenAI's enterprise agent, inside ChatGPT Enterprise. According to OpenAI, it is Uber's first company-wide rollout of an agentic platform built for knowledge work.
What happened
The original pilot ran across finance, operations and product. Uber's VP of engineering Sumanth Sukumar told TechCrunch that around 80% of the early adopters were non-engineers, using the agent to analyse operational data, localise training material for different markets and assemble dashboards to run their day.
The expanded deployment connects the agent to the tools Uber's teams already live in - Google Drive, Gmail, Google Calendar and Slack - so it can research, draft meeting briefs, build presentations and push work through multiple applications in a single run. GPT-6.1 Sol becomes the default model inside ChatGPT Work, with voice and image generation available when a task needs them. Engineering keeps Codex for coding workflows.
Why it matters for builders
Two details stand out. First, the deployment is supported by an internal AI skills library, an approved list of integrations and dedicated training - the unglamorous plumbing that decides whether an agent rollout survives contact with real employees. Second, Uber paired 30 AI engineers with business teams for two weeks and let them automate parts of their own jobs. Those "AI pods" produced the use cases that justified scaling.
Uber has not disclosed the spend and says it will judge the rollout by the workflows employees build, not by seat count. That is the correct test. Seat licences are the cheapest part of an agent programme; the expensive part is the identity, permission and integration layer underneath - the same bottleneck that has stalled agent deployments at retailers and banks all year.
The competitive read
The announcement lands while OpenAI works to defend its enterprise lead against Anthropic and as reports circulate about a new funding round at a $1.25 trillion valuation. Large reference customers matter more than model benchmarks in that fight: a 30,000-seat agent deployment is a public proof point that competitors must answer.
The practical signal for anyone building automation is simpler. The winning pattern here is not a single clever agent - it is a governed platform, a curated skills catalogue and a team that knows the job well enough to hand it over.
The Automation Brief
Read 5 AI stories instead of 50.
The essential moves in AI agents, models, automation and infrastructure — filtered for builders and operators, with the part that actually matters.
No noise. Unsubscribe anytime.
Editorial notes
Stefan Trbojevic
n8n Lab Editorial
23 October 2026
23 October 2026
Sources
AI disclosure: AI assisted with research and drafting. Factual claims are reviewed by an editor.
