A future-historical authority map · Edition MMXXVI

The Delegation Frontier

Where AI should advise, act, and stop

AI should get more autonomy only when evidence, permissions, reversibility, and accountability can keep ahead of the cost of failure. Capability alone is not enough.

These positions form an interpretive operating model for 2026, not a universal risk score. Facts, inferences, and speculative futures are labeled below.

View
Filter by consequence
I · The authority frontier

Autonomy must fall as consequences rise

Horizontal position shows how much autonomy AI has. Vertical position shows the consequence of a mistake. Four zones define the boundary: assist, approval required, bounded autonomy, and human-owned authority. Click or focus a work bubble to open its operating contract.

Position is interpretive · bubble size approximates operational frequency ◉ active 2026 frontier · bounded finance / operations work
I · The control ledger

The same work, ordered by consequence

The ledger removes the geometry and exposes the operating contract: who holds the decision right, and which control makes delegation revocable.

II · Calibration

Facts, inferences, and speculation

Evidence of capability does not grant authority. Each claim names its epistemic status and the reason for confidence.

III · Branching futures

Four ways the frontier could move

The next boundary will be set by evaluation, institutional control, regulation, and who remains accountable. Benchmark scores alone cannot set it.

IV · The recurring pattern

Delegation is a revocable control loop

A mature system widens authority only after people can observe the work, constrain it, test it, and reverse it.

The closing argument

A mature AI-native company can explain, test, and revoke every delegation. The amount it delegates matters less than the control it keeps.

Authority lasts only while control holds