Sansi, I'm back to bother you again haha.
New question this time: when you have an agent fix the frontend, how does it know the page actually came out right?
Right now I'm basically the agent's camera.
It ships a version, I screenshot it and describe what's broken, it ships another, I screenshot again...
I did look into browser-use, and it was a disaster.
To reuse my login state it had to hook into my Chrome, and then it kept spawning tabs nonstop.
I'd be reading a doc and the page would get yanked away mid-scroll.
I kept assuming I had something misconfigured.
It's not you. browser-use talks over CDP and takes over the entire browser — there's no "just borrow one tab" mode, so its automation collides with your everyday browsing.
It annoyed me enough that I switched to ego-lite.
What's ego-lite?
It's a browser built specifically for agents.
You let the AI use ego-lite as its browser, and you keep using Chrome.
So it gets one set and I get another — physically separated?
Right. After the agent changes the code, it opens the page in its own space, and if something looks off, it goes back and keeps fixing.
Oh right, what about login? When the agent opens our admin panel, surely it doesn't log in from scratch every time.
ego-lite can import your Chrome cookies and login sessions, so what the agent gets is already a logged-in state.
Much nicer than browser-use on that front.
Any pitfalls? Fair warning — I don't want to tinker all night and find out the time was all wasted.
Two I know of: one, it currently mostly supports Mac; the other is that on complex canvas-heavy pages, its snapshots are occasionally a bit off.
Mac is fine, I happen to be on a Mac.
Alright, I'll tinker with it this weekend. If I hit pitfalls, I'm coming back for advice.