The closest analogy I can think of is legal trusts, that have a stated purpose with little room for interpretation, but still has a human performing the operations.
Darknet markets take bitcoin, and at least some VPS providers do as well, so no need to get "real" bank accounts and evade know-your-customer.
Perhaps it could make and sell artworks.
However it's not theoretically impossible. Plenty of services are done over the internet with no human interaction.
Also, nobody says that such a bot wouldn't be profitable. It could send dividends to the creator (or do something else worthwhile with the money - a'la trust funds), it's just that the creator wouldn't be able to modify or stop it any more.
Right now it is an extremely concrete example, and really easy to say that the originator of the bot is at blame and should be prosecuted for buying illegal items.
But, how advanced does a bot have to be before it itself is at blame? What if they'd programmed it to reach out and purchase from any vendor it could find? What if it wasn't programmed to do anything but made random acts, took feedback and then learned from it?
That said, if you decide to keep any drugs you get from the not rather than immediately disposing of them, you've demonstrated the requisite intent to be guilty of possession, so there's that.