COMPUTER-USE AND BROWSER AGENTS

AGENTS THAT BROWSE AND ACT ON THE WEB A browser agent is an agent that uses a web browser to read pages and take actions on websites. Some browser agents work from screenshots like computer-use agents. Others read the structure of the page, which is the underlying text and the list of buttons, links, and form fields, often called the page content or the accessibility information. Reading structure is usually faster and more exact than looking at pictures, though some pages are hard to read this way. A browser agent can navigate to addresses, read visible text, fill in form fields, click buttons, scroll, and switch between tabs. Think of a very fast, very literal intern who can use any website but who does not know which sites to trust. Typical uses include searching for information across many sites, comparing prices, filling in repetitive forms, and collecting posted listings, such as job postings. The harness for a browser agent must handle several practical matters. Page loading: pages change while loading, so the agent must wait until the page is ready. Logins: pages behind a sign-in raise sensitive questions about credentials, covered below. Pop-ups and banners: cookie notices, advertisements, and chat boxes can block the real content. Different layouts: the same site may look different on different devices. Terms of use: many sites forbid automated access in their terms of service, and some use tools that detect and block automated visitors. Respect those rules, and do not try to defeat protections that a site has set up to stop automation, such as puzzles meant to prove a visitor is human. If a site offers an official API or a data feed, use that instead.