quality_assurance · travel · workflow
How Uber uses AI for development: inside look — Minion, Shepherd, uReview, and other internal agentic AI tools
Uber's traditional development workflow was single-threaded and manual, with engineers spending most of their time writing code themselves. Expanding AI tooling companywide proved harder than expected, with adoption slower than anticipated and AI-related costs rising sharply.
How it works
Common implementation structure
How this type of workflow is generally built, generalized across documented cases — not tied to any one vendor's stack. Click any stage to read what happens there. Specific products that implement these stages appear in “Tools commonly seen” below.
Stage 1 · Parallel agent kickoff
Engineers kick off multiple parallel background agents simultaneously while waiting for earlier agents to complete.
Tools used
MinionShepherduReviewCode InboxAutocoverClaude CodeGitHub CopilotCodexCursorIntelliJMichelangeloAIFX CLIUber Agent BuilderAgent StudioMCP gateway
Outcome
Uber achieved 92% monthly agent adoption among developers, with AI generating 65–72% of code inside IDEs and agents opening 11% of all pull requests. Engineers report higher satisfaction and are able to create features previously thought impossible.
Results
Time saved92%
Volume84%
Cost replaced100%
Grounding & classification
Source type: technical build writeup
48 fields verified against source quotes.
agentic workflowai agentcode generationmulti agent workflowquality inspectioncode diff prknowledge basemetric backednamed customerproduction runtime claimedsource backedtools describedworkflow describedsoftwaretravelautomation rateemployee productivitythroughput increasetechnical build writeupback office opsquality assuranceagentic task executionautonomous resolution