browser-use is a library that lets an AI agent open pages, click and fill in forms from a plain-language instruction. What it promises, and what to be clear about before using it.
Some tasks have no API.
Just a website, a form and someone losing an afternoon filling it in.
That's what browser-use is for.
browser-use is a Python library that connects a language model to a browser. You describe a task in words ("search for this", "fill in this form", "check this website and tell me what's broken") and the agent opens pages, clicks, types and reads the way a person would.
Under the hood it uses browser automation tools that already existed. What's new is that you don't have to program every step: the model decides what to do by looking at the page.
Because it reaches a part of work almost nobody has automated: what lives on other people's websites.
Supplier portals, old admin panels, government forms, sites without an API. All of that gets done by hand because programming a bot for each case isn't worth it. And when the site moves a button, the bot breaks.
An agent that understands the page by what it sees, not by its exact code, copes better with those changes. Not always. But better.
Automating isn't doing faster what you already do. It's no longer doing what should never have been done by hand.
This is where caution matters, because it's easy to get carried away.
For internal, repetitive, low-risk tasks. Checking your own website for errors. Testing that a form works. Gathering public information that someone currently copies by hand into a spreadsheet.
I'd start there. With the boring and the safe. And only later, very carefully, with the rest.
That the next frontier for AI isn't answering better, it's doing things. And the more a tool can do, the more it matters to decide carefully what you let it do.
P.S. Before automating a task with an agent, ask whether that task should exist at all. Sometimes the best automation is getting rid of it.