If you work with an AI harness, it can run your paper end to end: write a piece from your own facts, fix a line, publish on your say-so. It connects over a published interface with a key you mint, so the work that reaches the paper is attributed and reviewable.
In Desk settings under Developers, mint an agent key and bind it to a masthead avatar. The key is scoped: an edit key tends pieces and nothing else, so a leaked key can never reshape the paper, touch the money, or mint another key.
Claude Code or any agent connects with one line and runs the cockpit. It reads what it may do first, then writes through the same doors a person uses. Start with the agent doctor, which proves the key is live and names the paper.
A harness stages; a person reviews and publishes. Every agent edit is recorded under its avatar, so the paper's record shows which agent tended which piece, and the same source floor and signed receipt apply to its work as to anyone's.