AI agents can now be tested on real phone apps, not just fake ones

Researchers built a new way to test AI agents that control mobile phones. This lets developers see how their AI performs on real apps, not just simulated ones.

Until now, AI agents that control phones were tested in fake environments. They often looked good on paper but failed when faced with real apps like social media or ride-sharing. This new test uses actual third-party apps, which means developers can finally train AI agents that work in the messy real world.