The Trust-Autonomy Gap

Reader Toolkit

This is the page the book sends you to. The templates are listed below. The editable files are not posted yet.

Templates

The book names these working files. The page is here. The editable files are not posted yet.

  • Agent Authority Card

    An editable record of one agent: identity, purpose, tools, data, limits, owner, and how to revoke it. The reference card is Appendix F.

  • AI Safety Review Board charter

    A starting charter for a standing body with the decision rights to approve, constrain, suspend, and retire agents.

  • Delegation Dilemma worksheet

    A sheet for rating a task's consequence and complexity, then choosing the oversight that matches. The matrix below is the reference.

  • 90-day plan

    Three phases from the book: days 1 to 30, assess and assemble; days 31 to 60, architect and pilot; days 61 to 90, analyze and scale.

Choose the oversight before you delegate

The worksheet uses this matrix. Consequence and complexity pick the operating mode. A more capable agent does not, by itself, move a task into a lighter box.

Four-quadrant matrix mapping consequence and complexity to fully delegated, checkpointed, human-oversight, and human-in-the-loop operating modes.View the figure larger

The Delegation Dilemma Matrix

The matrix combines task consequence with task complexity to select an oversight model. Low-complexity, low-consequence work can be fully delegated and monitored by exception. High complexity with lower consequence calls for automated checkpoints, while high consequence with lower complexity requires pre-execution human oversight. Tasks that are both complex and consequential remain human-in-the-loop. The framework treats delegation as a revisable judgment rather than a permanent technology setting. Tasks should move between zones only when evidence, governance maturity, and the reliability of the relevant agent capability justify the change.

Takeaway. Oversight should increase with consequence and complexity, regardless of how capable the agent appears.

Continue

After publication

When a control, a standard, or an incident changes something I wrote, I will post the note on this page and send it to the list.