A browser AI agent, not a search box
Plenty of AI tools can search the web. OperatorNest’s operator can browse it: open a page, read it, click through to another, compare what it finds, and fill in a form field by field, using its own browser in its own cloud workspace. That’s a different, and more useful, kind of access than a search summary. Most real web tasks mean opening several pages, comparing what’s on them and getting something ready to submit or send.
What browsing does that a search summary can’t
Most of the web isn’t built for a single API call to summarize. Prices depend on a date you pick, availability depends on a form you fill in, and a fair comparison usually means visiting several pages and reading each one the way a person would. A browser AI agent can do that: click, scroll and read pages in the order the task requires, instead of returning a single snippet.
What a web task looks like
A few examples of what this covers in practice:
- “Compare business checking accounts from these four banks and tell me which has the lowest fees for our volume.” Your operator visits each site, reads the fee schedules, and builds a comparison.
- “Find three catering options for 40 people near the office on the 14th and get quotes.” It checks availability, fills in inquiry forms with your details, and reports back what each one said.
- “Fill in the vendor onboarding form with our company details from the shared doc.” It reads the source document, fills the form, and shows you the completed version before anything is submitted.
In each case, the browsing and the drafting happen on their own. What happens next depends on whether the step is consequential.
Where it stops and checks with you
Reading, comparing, and filling in a form are not the same as submitting it. Your operator will prepare a form completely, ready to go, and then stop before it submits anything that creates an account, pays, or confirms a booking. You see exactly what it’s about to send, and you approve, edit or hold it. See approvals for the full list of what always waits.
For example, after the catering research above, you’d see: “Ready to send inquiry forms to all three caterers with the details above. Approve to send?” Nothing goes out until you say so.
Handling logins and existing accounts
Some web tasks need access to an account you already have, like checking a booking or pulling a statement. Your operator uses only the access you’ve explicitly given it for that purpose, inside its own workspace, and nothing beyond what the task needs. It won’t create new accounts or change account settings on your behalf without the same approval step as any other consequential action.
Web tasks that run in the background
Because a web task can involve checking many pages or waiting on a form’s response, it often runs as a background task: you send it, close the chat, and it keeps working. A vendor comparison across a dozen sites doesn’t need you watching every page load; you’ll see the finished comparison when it’s ready, or a decision waiting if one came up along the way.
Web tasks and the channel you asked from
A web task doesn’t require you to sit inside a browser tab watching pages load. You can ask for a web task from WhatsApp, Slack, email or the web, close whatever you were doing, and get the comparison or the completed form back in that same place. See background tasks for how longer web tasks keep running after you step away.
A record of exactly what it did online
Because web tasks touch real sites and sometimes real forms, every web task’s receipt shows which pages it checked, what it filled in, and what it was waiting on you for. If a quote looks off or a form submission needs double-checking, you can see precisely what your operator read and entered, not only the end result. That’s the same level of accountability as any other task, applied to work on the open web instead of inside a single app.