[00:02]
browser. Your AI writes the code, but who checks if it actually works? Still you, right? Clicking around, refreshing pages, taking screenshots. Google just handed that job to the agent, and the rules it plays by are the part nobody's
[00:17]
reading. In this video, I'm going to show you what browser use in Google anti-gravity actually does, who can get it right now, and the one setting I'd check before you let it loose. Stay for that last part because it changes how
[00:30]
much you can trust it. Let's start simple. Anti-gravity is Google's agent-first coding platform. You tell an agent what you want, and it plans the work, writes the code, and runs it for you. It came out in November 2025, and
[00:43]
in May this year, it became a full desktop app where you can run lots of agents at once. Now, here's what's new. The anti-gravity team just posted that browser use has been rolling out in the app. They said it closes the loop
[00:56]
between researching, writing code, and checking it live. That means the agent doesn't just write your code and hand it back. It opens a real browser right inside the app in the side panel, and it uses it. It clicks buttons, it fills in
[01:09]
forms, it checks how your page looks on different screen sizes, it takes screenshots, it reads console errors and network requests, all the boring checking you used to do by hand. And here's the smart part. You don't need a
[01:21]
special command. If your task needs a live page, the main agent starts a browser helper on its own. Google calls it a browser sub-agent. If you just ask a simple question about some docs, it won't bother opening a browser at all.
[01:34]
And if you want to force it, you can type slash browser then your task. That browser helper runs in its own separate memory, so it doesn't clog up your main chat. By default, it uses a Gemini Flash model, and if your Flash limit runs out,
[01:47]
it doesn't quietly switch to something else. It stops and asks you what to do. You can also watch it work. When the agent takes over a tab, the tab icon turns into a little cursor arrow. Then an animated cursor moves around page
[02:01]
with a small label showing the step it's on or thinking while it plans. Okay, quick pause before I show you the next part. Inside the AI Profit Boardroom, we've got a 30-day roadmap that takes you through learning AI agent tools step
[02:14]
by step from your first setup to running agents on real tasks. There are also four coaching calls every week. If you get stuck on a setting, a permission prompt, or a task that just won't finish, you bring it to a call and work
[02:26]
through it with help. Everything you're seeing in this video, like the browser the kind of thing we break down together. Link in the description or go to aiprofitboardroom.com. All right, back to it. You can see this
[02:40]
in the demo video Google posted with the announcement. The agent is asked to audit the pricing page, the docs, and the changelog for a sample product. It says it's opening all three in parallel browser tabs and three separate tasks
[02:52]
start at once. Then you watch it work through the pricing page in the side panel while the chat shows each step. Now, the question everyone asks first, what about logging in? Here's how Google handles it. When the agent hits a login
[03:05]
page, like Google sign-in, a Microsoft or Okta login, a two-step code, a passkey, or one of those prove your human checks, it stops. It leaves the login page open and asks you to sign in. You type your details yourself in that
[03:19]
tab. The agent never enters your passwords. Then you reply something like signed in, continue, and it picks up right where it left off. That's what Google means by you choose what to log into. You decide which accounts it gets
[03:32]
to use, and those logins stick. The built-in browser keeps its own saved profile for each anti-gravity account, so you stay signed in even after you close tabs or restart the app. It's also kept separate from your normal Chrome
[03:45]
profile, so it isn't poking around in your everyday browsing. And the sign-in pages themselves are off limits for the agent to click around on. Next, downloads. This one landed in version 2.3.1 on November 9th. When the agent
[03:58]
tries to download a file, the download now pauses. You get a prompt asking if you want to allow it. It shows you the file name, where it came from, the file type, and which folder it'll save to. You can say yes for just this run, for
[04:11]
the whole conversation, or always for that website. And some files are never allowed. Installers, scripts, and app files like .exe, .dmg, and .sh files are always blocked in tabs the agent controls. Anything you do approve lands
[04:26]
in your downloads folder with a little agent badge so you know where it came from. Uploads are careful too. Every upload needs your okay every single time, and the agent can only upload files you've set up for it up to 20 at
[04:39]
once. Same idea with websites. If the agent tries to go to a site you haven't approved, it pauses and asks. You can allow it once, allow it for this chat, or always allow that site. And here's a sneaky detail I like. Local addresses
[04:53]
are checked by port, so approving localhost:3000 doesn't automatically approve localhost:8080. Big Google apps like Gmail, Docs, Drive, and Calendar need their own rules too. Approving one Google site doesn't open
[05:07]
the door to all of them, and pages the agent controls can't lock your mouse or keyboard or go full screen on you. So who can actually use this? The built-in browser agent is in the anti-gravity desktop app version 2.3.0 or newer on
[05:21]
Mac, Windows, and Linux. You need a personal Google account, and it's on the personal Google account, and it's on the free, plus, Google AI Pro, and Google AI Ultra plans. It's not on business or enterprise accounts, and it doesn't run
[05:33]
inside remote agents. Google says the rollout is still going, so if you don't see it yet, give it a little time. If you're on the older anti-gravity IDE, it's different. The IDE doesn't have a built-in browser. There you type
[05:47]
/browser and the agent drives your own local Google Chrome instead. That needs some setup. It shares your normal Chrome sign-ins, and it only runs one browser session at a time, so the app is the easier way in. Now the one setting I
[06:01]
promised. Go to Settings, then Browser. Find Browser Actuation Rules. This is where you set allow, ask, or deny for two things, which sites the agent can there's a site you never want it touching, put it on deny. If you want it
[06:17]
to check with you first, set it to ask. There's also a preset called Turbo that auto approves normal site visits and standard downloads. That's handy for speed, but sites on your deny list stay blocked and uploads still need your
[06:30]
okay. So set up your deny list first, then decide how much freedom to give it. choose where links open, change your download folder, and clear your browsing history. One last limit worth knowing. The built-in agent sticks to set actions
[06:45]
like clicking and typing. It can't run terminal commands or random code inside the page. It's built to check your work, not take over your computer. So here's the big picture. Before, your agent wrote the code and you were the tester.
[06:59]
Now it can write it, open it, click through it, and show you what happened while you stay in charge of logins, downloads, and where it's allowed to go. most business owners haven't figured this out yet, and that window won't stay
[07:13]
open forever. The AI Profit Boardroom helps you get ahead on learning tools like Anti-Gravity. You get daily tutorials, so new updates like this browser agent get explained step by step. You get four coaching calls every
[07:25]
week where you can bring your questions about setup and settings. There's a prompt library, so you never start from a blank chat box. And there's the member are and connect with them. Link in the description or go to
[07:38]
aiprofitboardroom.com. If you want the full process, the SOPs, and over 100 AI use cases like this one, join the AI Success Lab. It's our free community. Links in the comments and description. You'll get all the video
[07:51]
notes from this episode plus access to over 94,000 members. It's a place to learn AI alongside people figuring it out too. The video notes pull together the key points from this episode, like where the browser settings live and how
[08:04]
the approval prompts work, so you don't have to rewatch to find them. It's free to join. So come in, grab the notes and I'll see you inside.