The fastest way to lose trust in an AI that runs your marketing is one post you never saw. The second fastest is an AI that asks about everything. Mother sits between the two by sorting every action she can take into a class, and letting the class decide what consent it needs.
Four classes
| Class | For example | What it needs |
|---|---|---|
| 0 · Inside Mother | Drafts, campaigns, schedule changes | Nothing. Undo, and a 24-hour restore. |
| 1 · Spend | Image and video generation | A budget per brand. |
| 2 · Outside, can be withdrawn | Scheduling a post, a social reply | Your clear instruction, or your approval. |
| 3 · Outside, final or money | Publishing now, sending to a list, ad spend | Your clear instruction with 30 seconds to cancel, or your approval. |
A clear instruction is a yes
When your own message names the action (“schedule this for Friday on Instagram”), that is the approval, and Mother doesn’t ask again. Anything less clear becomes an approval card, where the work lives: replies in the Inbox, content in Content. Each card carries the time it is needed by, so nothing quietly expires.
Asking less, with evidence
Autonomy climbs one class at a time: draft, then batch approve, then exceptions only, then on her own within a budget. Mother proposes each step with the evidence (“47 of the last 50 replies approved unchanged”), you decide, any incident takes it back, and a random sample stays audited.
Autonomy is earned per kind of action. It is never a switch.
The classes and approvals with deadlines are in Mother today. The ladder that lets her propose her own promotions is coming in 1.2.