KAHuman Control Plane

Interactive product study · Enterprise AI

Decide where AI stops.

Change the operating conditions. The control policy updates before the agent gets authority.

Capability is not permission.Authority follows consequence, reversibility, evidence, and intent.
Scenario
02

Recommended control policy

Fresh approval

Policy recommendation

Preview + fresh approval

Pause before the consequential action and bind approval to the exact committed state.

ConsequentialThe cost of error is material.
Hard to reverseRecovery cannot be assumed.
Intent can expireApproval must match current state.

Agent may

  • Fill reusable profile facts
  • Draft role-specific answers
  • Validate required fields

Human owns

  • Review subjective answers
  • Resolve salary and sponsorship
  • Issue a fresh submit instruction
Current statePolicy ready
RUN-0042

Prepared form is inspectableNo external action has occurred.

Reusable product patterns

Four controls, not one approval modal.

The right pattern changes with the action. Each card loads a scenario that makes the boundary concrete.

Product judgment

The operating model is part of the interface.

01

Match authority to consequence

A model's capability does not grant it permission. The product assigns authority based on potential harm.

02

Bind approval to state

Approval expires when inputs, evidence, navigation, or the proposed action changes.

03

Design recovery with execution

Undo, reject, escalate, and fail-closed states belong in the primary workflow.