Quality assurance · Production

How Uber uses AI for development: inside look — Minion, Shepherd, uReview, and other internal agentic AI tools

The problem

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.

Workflow diagram · grounded in source
1
Parallel agent kickoff
trigger
“engineers get into this mode of running several agents at once”
2
Internal context retrieval
integration
“accessing Uber's source code, engineering documentation, Slack information, JIRA tickets, etc. These all serve as "memory" for agents to use”
3
Minion background execution
ai_action
“Uber built Minion, an internal background agent platform with monorepo access and optimized defaults”
4
Autocover test generation
ai_action
“Autocover for generating 5,000+ unit tests per month”
5
Code Inbox PR routing
routing
“Uber built Code Inbox for smart PR routing”
6
uReview AI code review
validation
“uReview for high-signal AI code review comments”
7
Shepherd migration management
ai_action
“Shepherd for managing large-scale migrations end to end”
Reported 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.

Reported metrics
agentic coding users among Uber devs84%
code AI-generated inside IDE-based tools65-72%
code AI-generated in CLI tools like Claude Code100%
Claude Code adoption baseline (December 2025)32%
Show all 10 reported metrics
agentic coding users among Uber devs84%
code AI-generated inside IDE-based tools65-72%
code AI-generated in CLI tools like Claude Code100%
Claude Code adoption baseline (December 2025)32%
Claude Code adoption (February 2026)63%
Uber devs using agents monthly92%
pull requests opened by agents11%
AI-related cost increase since 20246x
unit tests generated per month (Autocover)5,000+
engineer satisfactionmuch higher satisfaction from our engineers
Reported stack
MinionShepherduReviewCode InboxAutocoverClaude CodeGitHub CopilotCodexCursorIntelliJMichelangeloAIFX CLIUber Agent BuilderAgent StudioMCP gatewaySlackJIRA
Source
https://newsletter.pragmaticengineer.com/p/how-uber-uses-ai-for-development
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

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.

What tools did this team use?

Minion, Shepherd, uReview, Code Inbox, Autocover, Claude Code, GitHub Copilot, Codex, Cursor, IntelliJ.

What results were reported?

agentic coding users among Uber devs: 84%; code AI-generated inside IDE-based tools: 65-72%; code AI-generated in CLI tools like Claude Code: 100%; Claude Code adoption baseline (December 2025): 32% (source-reported, not independently verified).

How is this quality assurance AI workflow structured?

Parallel agent kickoff → Internal context retrieval → Minion background execution → Autocover test generation → Code Inbox PR routing → uReview AI code review → Shepherd migration management.

Related quality assurance cases