Google's Gemini just started actually using your apps for you, and the agent economy stopped being theoretical.
The Signal
This isn't another chatbot that spits out answers. Gemini now opens your food delivery app, navigates the interface, selects items, and completes the transaction. It calls you a rideshare. It operates in what Google calls a "virtual window," meaning the AI is literally manipulating app interfaces the way a human would, tapping buttons and filling forms while you watch.
This matters because it solves the integration problem that's been choking AI assistants since Siri launched in 2011. Previous assistants needed explicit API partnerships with every service. DoorDash had to build special hooks for Alexa. Uber had to create a Google Assistant integration. The AI could only do what developers explicitly programmed it to do. Gemini sidesteps all of that. It just uses the app. No special API needed. If you can tap it, Gemini can tap it.
The beta launched on Samsung's S26 Ultra and Google's Pixel 10, starting with food delivery and rideshare. That's a narrow wedge, but the wedge matters more than the width. Once an AI can operate one category of apps through interface manipulation, it can theoretically operate any app with a graphical interface. Banking apps. Healthcare portals. Your company's internal tools. The constraint isn't technical capability anymore. It's trust, liability, and whether people actually want their phone operating itself.
The Implication
Watch how Google expands the app categories. If they move into financial services or healthcare within six months, they've solved the liability puzzle and this becomes the default interface layer for mobile computing. If they stay narrow, something spooked them. Either way, every SaaS company should be gaming out what happens when users stop opening their apps directly.