See the page, not just the code
An agent that can't see the page it changed is guessing. In ace, agents open pages in a real Chromium browser, read the page as structured text, click, type and scroll, and check the console and network log for errors, then fix what they find.
The tab is shared: you watch the agent work in it, and you can take over whenever you like. Screenshots at phone, tablet and desktop widths let an agent check a responsive layout in one step.
Watch it from anywhere
The browser agents use is one shared session, not a copy per viewer. Open it on your phone, in the web app or on another desktop and it streams live at 60 frames a second, and you can click and type in it. Open a tab on the phone and the page switches to a phone-sized layout, and the agent is told it is working at that size.
Point at what's wrong
Mark up a page directly, circling the broken button or the misaligned card, and send the notes to the agent with the page attached. It is the fastest way to say exactly what you mean about a UI.
Your own Chrome, when you need it
Some work needs your real sessions. Use my Chrome lets agents work in your own Chrome profile through an extension, but only in tabs you share from it or tabs they open in a separate ace window. Chrome shows its debugging bar while they do, and Stop or Revoke ends it at once.
Settings pages, extension pages and the password manager are refused, cookie reads are blocked, and page storage reads are off unless you allow them. It works with Chrome, Chromium, Edge, Brave, Vivaldi and Arc.