What browser work would you actually hand to an AI agent?

I built WebCLI because agents can't really use the web to do useful tasks like drive cloud consoles, SaaS dashboards, vendor portals, complex forms and settings pages, admin panels, checkouts, etc. Building fragile brittle selector automations takes too long, I don't want to babysit, I just want to throw an agent at it and get on to something else.

WebCLI gives agents a local CLI + skill for driving real browsers (you can watch, with headful, that's the main use case). Human handoff for login, MFA, CAPTCHA, payment, or confusion.

I'm curious:

• What browser tasks would you trust an agent to handle today?

• Where do scripts break for you, but an adaptive agent could succeed?

Any real examples would be interesting and useful.