We started with a simple observation. LLMs are very good at writing code, but once that code runs, they lose visibility. They cannot see what is happening inside the app, and they cannot act on it. The feedback loop breaks. So we asked a different question. What if agents could interact with the React Native Debugger the same way a developer does? What if they could inspect the UI, follow state changes, read logs, and understand network activity in real time? And what if we went one step further and gave them hands to actually interact with the app? As we explored this, something shifted. Agents stopped guessing. They began navigating flows, reproducing bugs, and validating behavior on their own. Instead of generating code and waiting, they could iterate. In this talk, I will show what changes when agents can both observe and act inside a running React Native app, and how this unlocks a new kind of development workflow where agents do not just write code, but actively work with it.