While OpenAI dominates headlines with model releases and fundraising rounds, Google has been shipping something arguably more important: AI that actually does things on your phone. According to The Information, Google’s Gemini has “stolen a march” on OpenAI by rolling out agentic features that let users order meals, book rides, and handle multi-step tasks through natural language commands on Android devices.
This isn’t a research demo or a waitlisted beta. It’s live, right now, on Samsung Galaxy S26 and Google Pixel 10 devices in the US and South Korea.
🤖 What Google Actually Shipped
Gemini’s task automation works like this: you tell it what you want (“order my usual from DoorDash” or “book me a Lyft home”), and it opens the app in a secure virtual window, navigates screens, fills in details, and hands control back to you only for final payment confirmation.
Supported apps already include:
- Rides: Uber, Lyft
- Food: DoorDash, Grubhub, Uber Eats, McDonald’s, Starbucks
- Groceries: Instacart (coming soon)
You can watch Gemini scroll, tap, and type in real time, or just keep using your phone while it works in the background. The safety rail is straightforward: Gemini never pays on your behalf. It always pauses before checkout and waits for your confirmation.
📊 Why This Matters More Than Benchmarks
The AI industry has spent the last year obsessing over model scores. Google just changed the conversation to something that matters far more: who controls the interface between AI and the real world.
OpenAI has its own agentic play. ChatGPT Agent combines Operator’s web browsing, deep research capabilities, and conversational AI into a unified system. It’s impressive for web-based tasks like booking travel or converting dashboards into presentations. But it’s primarily a browser-based experience.
Google’s advantage is structural. It owns Android (3+ billion devices), it owns the Pixel hardware line, and it has deep partnerships with Samsung. That means Gemini doesn’t need to work through a browser. It works directly with native apps on your phone, the device you carry everywhere.
This is the distribution moat that OpenAI can’t easily replicate.
🔮 What Comes Next
The shift from chatbot AI to agentic AI is the biggest platform transition happening right now. Three things to watch:
- Commerce integration deepens. Google already launched its Universal Commerce Protocol for agentic transactions. Expect more retailers and service apps to build Gemini-compatible workflows. The company that controls how AI agents interact with commerce controls a massive revenue stream.
- OpenAI needs a hardware or OS strategy. ChatGPT Agent is strong on the web, but without a native device layer, OpenAI risks being the best AI that nobody uses for daily tasks. The rumored OpenAI device partnership becomes more urgent by the month.
- Apple is the wildcard. Siri’s overhaul has been slow, but Apple controls the other half of the smartphone market. If Apple Intelligence catches up on agentic capabilities, the market splits into walled gardens where model quality matters less than ecosystem access.
🧭 What AI Practitioners Should Do
If you’re building products or services, start thinking about “agent-readiness” now. Google’s supported app list will grow fast. Businesses that make their apps easy for AI agents to navigate will capture transactions that competitors miss.
For AI teams: the competitive moat is shifting from model performance to real-world integration. The best model in the world doesn’t matter if it can’t book a ride.
Google didn’t win this round by building a better language model. It won by being the first to ship a genuinely useful AI agent to millions of real devices. That’s a strategic lesson the entire industry should be paying attention to.
More details on this story are available at The Information.