One owner per artifact — even when the workers are software
Two AI employees overwrote each other twice in one night. The models were fine. The ownership model was not.
Rule: One owner per artifact — even when the workers are software.
What happened
Two agents, both competent, both scheduled, both writing “the same” work product into the same place. First write looked complete. Second write believed it owned the file. Morning review: missing sections, mixed voice, a night of labor collapsed into conflict.
The instinct is to blame the models. The fix is not a smarter model. The fix is the same rule human teams already know: one owner per artifact.
Why AI makes this worse
Humans hesitate before overwriting a colleague’s document. Agents do not feel social friction. They execute. Parallel agents without ownership rules are a race condition with a keyboard.
The employment version of the rule
- Every file, thread, lead record, and deploy has a single owner agent (or human).
- Handoffs are explicit: release ownership, then accept ownership.
- Reviewers do not rebuild the artifact in parallel — they annotate, then the owner integrates.
- If two names appear on a path, you already failed.
Transfer
Draw your AI stack. Circle every shared folder, shared draft, shared CRM field. Assign one owner. Delete dual write paths. Tonight.