I'm somewhat with the parent on this. Getting this right will probably take a long time but can eventually bring us something like J.A.R.V.I.S, whereby an AI ag

I'm somewhat with the parent on this. Getting this right will probably take a long time but can eventually bring us something like J.A.R.V.I.S, whereby an AI agent can reason about all activities you've performed on the computer in the past and give you advice or perform full tasks for you based on that. We can argue whether we want to give the AI that level of trust, but I'm very interested in the potential.