Cloud agents that test mobile apps on real devices
summary
Claims a runnable QA agent that exercises real iOS/Android simulators, captures screenshots to Vercel Blob, and comments evidence on PRs; built with npx eve@latest, an AI SDK + an agent-device tool, CI (Expo/EAS or GitHub Actions) and Vercel Connect for secrets. Demo runs in ~2–3 minutes, costs about $0.10 per run, and uses swappable models (shown with GPT-5 4 0.4 mini); it detects UI flow breaks like permission prompts, navigation loops, and release-only crashes and produces repeatable, bounded checks per PR.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:26AI agents in mobile dev
- 3:21Agent-driven code review
- 6:42End-to-end tests limits
- 8:50Mobile QA agents
- 9:58QA agent stack
- 13:11Live demo
- 16:06Five rules for agents
- 17:30Future of agent-driven dev
related talks
15mYour agents can run, now what?
40mDigitizing Testers
23mGiving agents hands and eyes in React Native
49mPanel Discussion: AI in Mobile QA - What Actually Works Today?
19mThe GPT-ification of Mobile App Testing
18mReinventing App Testing with Agentic Apps
46mBringing Mobile App Validation into the Agentic Loop
11mDragonCrawl: Revolutionizing Mobile Testing with GenAI