What Uber’s Genie teaches us about a company brain.
Less waiting for answers. Fewer repeated questions. A practical lesson in making company knowledge useful.
Uber’s engineers were asking roughly 45,000 support questions a month. Many answers already existed, but finding them meant searching scattered documentation or waiting for a colleague. Genie brought those answers into Slack.
Connect the knowledge to the working day
As ByteByteGo explains, Genie searches internal documents before generating a response. Relevant passages and their source links give the model evidence to use, rather than asking it to rely on general knowledge.
Employees can ask follow-up questions, give feedback or escalate to a person. Uber also evaluates answers and the documents behind them: poor source material limits the quality of the response. Sources were deliberately selected for appropriate access; connecting everything indiscriminately was not the approach.
Useful does not mean infallible
In its October 2024 account, Uber reported deployment across 154 Slack channels, over 70,000 questions answered and an estimated 13,000 engineering hours saved. It reported a 48.9% helpfulness rate—not a claim that every answer was correct or every issue resolved.
These are Uber’s historical estimates, not Ontara results or a promised return.
The lesson for a smaller company
Our takeaway: you do not need Uber’s scale to recognise the problem. A dispatcher needs a customer’s delivery rules. A new colleague needs the right procedure. A sales team needs to know what was promised.
Start with one recurring question and the people who answer it. Connect approved sources, teach the team to check the answer, and keep a clear route to human help. Measure whether people spend less time searching and repeating themselves—not simply how many messages the system produces.
Take it into the next conversation.
A company brain earns its place when people can find a trustworthy answer without interrupting someone else’s work.
Sources & further reading
ByteByteGo: How Uber Built a Genie to Answer Questions Correctly 19 October 2026. The reader-supplied explainer; technical details and reported results checked against Uber’s original account.
Uber Engineering: Genie: Uber’s Gen AI On-Call Copilot 10 October 2024. Historical, self-reported deployment results—not current usage or a forecast for another company.
Source pages checked 19 October 2026. Our examples and recommendations are editorial interpretation, not findings from those publications.At Ontara we believe that every company deserves this.
Talk to us