DOCS · CONTROL

    Human approval, configured authority, and AI action.

    A practical explanation of prepared work, user-invoked actions, standing workflow configuration, and where control can vary.

    Reviewed 31 July 2026
    IN BRIEF

    AUGMTD separates work preparation from consequential execution, but approval behavior varies by action and configuration. Ordinary suggested replies are reviewed and sent by the user. Internal organization can occur automatically. Configured workflows may carry standing authority for specific external delivery or tool steps.

    01

    Preparation can happen before review

    Draft replies, nudges, documents, meeting follow-ups, briefs, and plans can be created before the user opens the relevant surface. Preparation does not itself mean an external action occurred.

    02

    User-invoked actions remain explicit

    Ordinary sends and many consequential actions appear in product flows where a user reviews or initiates execution.

    03

    Configured workflows can carry standing authority

    A scheduled workflow may deliver a report or perform an enabled tool step when a workspace has deliberately configured that behavior. This is different from a per-run approval.

    04

    The interface should show the authority model

    A useful control surface identifies whether an item is a suggestion, a draft, an approved step, a configured automation, or a completed external action.

    05

    Review quality matters

    A reviewer needs enough context, time, authority, and evidence to recognize a weak output. A confirmation button alone does not create meaningful oversight.

    DECISION GUIDE

    Choose by the shape of the work.

    Prepare only

    Use for work that benefits from context assembly and drafting while preserving a mandatory human decision.

    User-invoked execution

    Use when the action is consequential and should occur only after a person starts the step.

    Configured recurring delivery

    Use for bounded, repeatable work where the owner understands the tools, recipients, schedule, and failure handling.

    LIMITS AND CONDITIONS

    Where the boundaries sit.

    • AUGMTD does not currently use one universal approval mechanism for every tool and action.
    • A blanket claim that nothing is ever sent without per-run approval would be inaccurate.
    SOURCES AND FURTHER READINGWhy human oversight does not automatically improve AI workEvidence-led analysis