Does Chromium now include the infrastructure for AI “Agent Mode” (without the model itself)?

[Lawrence Imabr]

Hi everyone —

I’ve been following Chrome’s recent announcements about Gemini 2.5 and “AI Agent Mode” / “computer use” features. I understand that the Gemini model itself is hosted and proprietary, so it’s not part of Chromium.

What I’m trying to clarify is whether the infrastructure and control logic for that agent functionality, things like orchestration hooks, browser-control APIs, permission gating, or task interfaces, are now part of the open-source Chromium codebase.

In other words, does Chromium already include the plumbing needed for a connected AI model to inspect pages, take actions (e.g., navigate, click, fill forms), or run agentic tasks once connected to a backend model?

Or are those capabilities still confined to Google’s internal Chrome builds and not yet merged into Chromium?

If anyone has insight into whether this functionality exists in current Chromium branches (e.g., DevTools MCP integration, experimental flags, or new browser automation APIs), I’d love to know where that work lives or if it’s planned for later integration.

Thanks!

- Lawrence Barmerio