Why an AI operating system needs a native browser
The web is where much of the world's computer work happens. For a persistent agent, the browser cannot remain a borrowed window with temporary state.
A native browser gives an agent a durable, structured, governed way to use the web. It exposes what the page is, what actions are possible, which identity is active, and what evidence an action produced.
- Preserve sessions and task context across work.
- Expose semantic structure without forcing every action through pixels.
- Keep visual rendering available for compatibility and verification.
- Connect credentials, permissions, downloads, and evidence to the resident.
The browser is the real application layer
For many people, the browser already is the computer. Messages, documents, banking, shopping, support, research, administration, and internal tools all arrive as websites. An agent expected to carry computer work needs the browser as a central system surface, not a fragile accessory.
Today an agent often enters the web through one of two narrow doors. It receives text extracted from a page, losing interaction and visual context. Or it controls a human browser through screenshots and pointer actions, regaining compatibility but paying an interpretation cost on every step.
Both methods are useful. Neither alone gives a persistent resident the full environment it needs.
Direct structure plus visual truth
Native does not mean replacing the web with a private agent protocol. It means the operating system understands the browser as part of the agent's world. Page structure, navigation state, available actions, active account, downloads, and history can be represented directly.
The visual page still matters. Some sites expose meaning only through layout, canvas rendering, images, or behavior. Screenshots are also valuable evidence for what a person would have seen. The native design should combine semantic access with visual verification instead of declaring one universal interface.
This also supports recovery. If a workflow breaks after five steps, the system can retain the page, session, prior actions, downloaded artifacts, and intended next state. The agent should not have to rediscover the entire route from a fresh tab.
The browser carries authority
Web actions can spend money, send messages, accept terms, change records, and expose private data. A native browser therefore cannot be only an automation engine. It must participate in permission and evidence systems.
The operating system should know which resident requested the navigation, which account is active, whether the action fits current authority, and which step needs approval. Afterward, it should retain enough evidence to explain what changed.
In ERIKA, the browser belongs beside memory, files, accounts, permissions, and recovery because those systems constrain one another. A downloaded contract becomes a file with provenance. A login is an account available under rules. A completed form becomes evidence attached to a goal.
A browser built this way is more than the place an agent visits the web. It is where the operating system turns an open, human-shaped network into a workable environment for a persistent resident.