Control
Action Plan
The agent proposes its plan and waits for approval before touching anything.
established named by shapeof.ai
Named by Shape of AI; now the load-bearing consent mechanism of the agent era. The plan is simultaneously a preview, a permission dialog, and a contract — the user edits intent before cost is incurred rather than repairing damage after.
Why it works. Approval of a plan is cheap; reversal of an action is expensive. Moving the human checkpoint upstream converts oversight from cleanup to steering.
Steal: make the plan editable, not just approvable — that’s the difference between consent and steering. Skip: plan-approval for trivial tasks; friction where the blast radius is zero teaches users to rubber-stamp.