I certainly see a lot of use cases for this, but shallow ones. For most things I would wish to have done I must login to the site. If I read the demo right,

I certainly see a lot of use cases for this, but shallow ones. For most things I would wish to have done I must login to the site. If I read the demo right, there is a browser that OpenAI runs that the agent is acting on, not the local browser on my computer. (which could be logged in) "Agent book me a flight Boston tomorrow before 11am"