Generative Ai For High Quality Mobile Testing

Uber built DragonCrawl to solve a problem familiar to any team running mobile end-to-end tests: script-based testers break every time a UI changes, and keeping them working had eaten 30-40% of engineers' time. DragonCrawl replaces brittle selectors with an LLM that looks at a screen the way a person would and decides what to tap next, so a redesigned button or a shuffled menu doesn't require someone to go rewrite a test script. The approach scales across more than 50 languages without separate localized test suites, since the model is reading the screen rather than matching hardcoded strings. It's a useful data point for anyone weighing whether to bring LLM-driven testing into a large, multilingual mobile codebase: Uber is reporting this in production, not as a research prototype, and framing the win specifically in engineer-hours reclaimed rather than abstract accuracy numbers. That's the kind of ROI case that tends to travel well internally when a team is deciding whether to invest in agentic test tooling.