Quality assurance · Production

Uber scales from 2 to 500+ Claude AI skills in 5 months through grassroots engineering adoption

The problem

Uber needed to move its entire SDLC toward agentic engineering across 200+ microservices and thousands of globally distributed engineers, while avoiding the top-down AI mandate pattern that has caused adoption to stall at other companies.

Workflow diagram · grounded in source

Skill loaded from Golden Marketplace

trigger

“Auto-loaded for all engineers, works out of the box”

Claude skill executes engineering task

ai_action

“a skill spins up iOS simulators, toggles dark/light mode, switches languages, and runs tests to confirm generated code doesn't break the UI”

LLM-as-a-Judge validates new skill

validation

“Before any skill ships to the Golden Marketplace, a second AI runs a battery of tests against it, checking outputs against expected baselines”

Engineer final review

human_review

“Engineers stay as the final reviewer. The AI shows its work.”

Deterministic output report produced

output

“Enterprise skills must report exactly what they attempted, what succeeded, what failed, and the exact diff”

Reported outcome

A single engineer's side project grew organically into 500+ specialized AI skills powering Uber's engineering org, with 200+ curated skills in a governed Golden Marketplace and 300+ experimental tools in team repos, with twenty new skills being added per week.

Reported metrics

total AI skills deployed500+

curated skills in Golden Marketplace200+

Experimental tools in team repos300+

New skills added per weekTwenty

Show all 5 reported metrics

total AI skills deployed500+

curated skills in Golden Marketplace200+

experimental tools in team repos300+

new skills added per weekTwenty

task automation time savingstasks that used to eat hours

Reported stack

ClaudeClaude CodeLLM-as-a-JudgeiOS simulatorsGitHubCI/CD

Source

https://medium.com/activated-thinker/how-uber-secretly-scaled-ai-from-2-to-500-skills-in-5-months-without-a-strategy-25ff894c0f9c

Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

A single engineer's side project grew organically into 500+ specialized AI skills powering Uber's engineering org, with 200+ curated skills in a governed Golden Marketplace and 300+ experimental tools in team repos, w…

What tools did this team use?

Claude, Claude Code, LLM-as-a-Judge, iOS simulators, GitHub, CI/CD.

What results were reported?

total AI skills deployed: 500+; curated skills in Golden Marketplace: 200+; Experimental tools in team repos: 300+; New skills added per week: Twenty (source-reported, not independently verified).

How is this quality assurance AI workflow structured?

Skill loaded from Golden Marketplace → Claude skill executes engineering task → LLM-as-a-Judge validates new skill → Engineer final review → Deterministic output report produced.