Ask HN: Anyone actually doing AI browser automation?

We've been experimenting with various browser use tools that are out in the wild. They all promise the same thing...natural language prompting for browser use automation.

Ignoring the typical challenges around Auth, 2FA, etc. Is anyone having success getting these out of "demo" stage and actually deployed with any scale?

I keep running into the same issues. Same prompt, same website, same input data, different results on screen (missed a click, wasn't focused on the right thing, flubbed the EU cookie modal, etc).

So I'm looking to poll the community here. Has anyone gotten this reliable to a point where they deploy it into the cloud and it runs without a human babysitter? If so, can you share some of your approach with us?