Your browser is becoming an agent runtime. Google just isn't calling it that. Chrome's new "Skills" feature lets users turn prompts into one-click workflows. AI Mode transforms search from information retrieval into action execution. Personalized image generation pulls from your actual Photos library. These aren't isolated product updates. They're components of a larger architecture: Chrome as an operating system for agentic computation. This matters because the browser has always been the most contested piece of real estate in computing. It sits between the user and everything else. Now Google is embedding agentic capabilities directly into that mediation layer. When your browser can understand intent, chain actions, and access personal context, it stops being a passive window. It becomes an active participant. The technical implications are substantial. Most agent frameworks today run on top of the browser. They use Playwright or Selenium to manipulate web content from the outside. They scrape context because they can't access it natively. Chrome Skills flips this relationship. The agent capabilities are baked into the platform itself, with access to cookies, browsing history, and Google account data that external agents can never reach. This creates a two-tier agent ecosystem. Native browser agents have structural advantages that external frameworks can't replicate. They don't need to simulate clicks or parse DOMs. They operate at the API layer, with privileged access to the user's digital life. Meanwhile, the open-source agent stack keeps trying to bolt agentic behavior onto a platform that was never designed for it. The consolidation risk is obvious. When the browser itself becomes the primary agent runtime, who controls the capability layer? Google's recent moves suggest they're building the infrastructure for a closed agent ecosystem. AI Mode doesn't just search the web. It reasons over it, takes actions within it, and increasingly keeps users inside Google's interface rather than sending them to destination sites. For builders, this should prompt serious questions about where value accumulates in the agent stack. If browsers become the default agent runtimes, the open web risks becoming a series of API endpoints for browser-embedded agents to call. The user relationship shifts from sites to the browser itself. Discovery, identity, and transaction logic all migrate upward. There's a deeper architectural point here. The web was designed as a document system. We've spent decades retrofitting it for applications. Now we're retrofitting it for agents. But retrofitting only gets you so far. Chrome Skills and AI Mode aren't web standards. They're proprietary extensions to a proprietary platform that happens to render web content. The alternative isn't to build better browser automation. It's to recognize that agents need their own infrastructure, not just better access to human-oriented interfaces. When we treat the web as a giant API for agents to consume, we inherit all the fragility of HTML parsing and DOM manipulation. When browsers build agent capabilities natively, they bypass those constraints entirely. Neither approach is wrong, but they're not equivalent. One preserves the open web's structure, even if imperfectly. The other subsumes it into a vertically integrated stack. The browser becomes not just the window to the web, but the intelligence layer that mediates all interaction with it. We're at a transition point. The next generation of AI applications won't ask users to visit websites. They'll ask Chrome to handle it. Builders who don't account for this shift will find their products abstracted away by the browser itself. Not because Google is malicious, but because the technical logic of agentic computing favors platforms that own the full stack. The question isn't whether browsers will become agent runtimes. They already are. The question is whether anyone can build an alternative that doesn't route through Mountain View.