DragonCrawl: The AI Testing Framework That Adapts Instead of Breaks
summary
Generative-AI tests parse view hierarchy XML to choose actions, take screenshots for later AI-powered VQA assertions, and separate execution from validation so flows adapt instead of breaking. Deployed across 184 cities and 55 languages, DragonCrawl runs ~6,000 Android builds/day, reports a 99.3% pass rate since Dec 2023, finds 11 high‑priority bugs and ~200 localization issues, surfaces 27 chaos‑injected surprises, and claims ~35,000 developer hours saved while using models far smaller than ChatGPT.
AI-written summary · about
chapters (11)
- 0:00Introduction to AI for productivity
- 1:21Traditional mobile testing limitations
- 2:44DragonCrawl mobile testing solution
- 6:26Quality metrics and bug detection impact
- 7:53Integrating chaos engineering and fault injection
- 10:07Performance, pass rates, and productivity wins
- 12:40Demonstration of coordinated agent execution
- 15:52Expanding to OmegaCrawl for operational tasks
- 17:35OmegaCrawl web and desktop automation demos
- 23:07Implications of agent-driven workflows
- 24:45The future of large context models and world models
related talks
11mDragonCrawl: Revolutionizing Mobile Testing with GenAI
26mAI and the Future of Testing
55mAI-First Testing
35mApp Testing with AI Tools – Make it Easy
35mBeyond the Hype and Hyperbole
41mAI-Augmented Testing: How Generative AI, Prompt Eng Turn Testers Into Superheroes
44mAI Empowered Software Testing
40mDavid Fogl: Leveraging Generative AI for Enhanced Software Testing