Browser Agent

A browser agent is computer use narrowed to the browser, which makes it considerably more tractable: web pages have structure an agent can read directly rather than inferring from pixels, and the action space is bounded — click, type, scroll, navigate.

The use cases are the tedious multi-step errands: comparing options across several sites, extracting data from pages with no API, filling repetitive forms, monitoring for changes. Vendors have shipped these both as standalone agents and as capabilities inside browsers themselves.

Two frictions are structural rather than temporary. Sites increasingly detect and block automated access, so agents encounter bot checks a human would not. And the web's economics assume human viewers — an agent that reads a page without seeing its ads is a cost to the publisher, which is why access terms for AI agents are becoming contested.

Browser agents are the most likely route by which agentic AI reaches ordinary consumers, since the browser is where most people's tasks already live. They are also the sharpest case of indirect prompt injection: any page the agent visits can contain instructions aimed at it.

Use them for retrieval and comparison; be cautious about anything transactional. Watch what an agent is authenticated as — one running in your logged-in browser can act as you on every site you're signed into.