Left alone, it drifts. A tool built to recommend ends up deciding and acting, and nobody held the promotion meeting.
Software used to carry out steps people had already chosen. Today's AI agents can recommend, decide and act. That changes the first question from "what can it do?" to "what are we actually handing over?"
Every key decision gets a rung, a written grant and one named sponsor who can see it, stop it and refuse it.
Ladder · Grant · Sponsor of record
People share one language, and the work runs in order: foundation first, automation last.
Five stages · Role-based training
Gains are counted on two ledgers, the enterprise's and its people's, with evidence for each.
Dual Ledger · Cost per accepted outcome
Enforced where the action happens. The Consequence Layer puts a gate between what an agent proposes and what actually occurs, owned by named people and proven by tests.
One ladder, six rungs, numbered zero to five. Each rung is granted for one particular decision, never stamped on a whole agent. Scroll through it, or tap a rung.
Ask: which rung is this on, and who put it there?
Before any machine touches the work, a person decides whether a machine belongs there at all. Some work stays here on purpose.
The machine does exactly what it's told. A person has already made the call; the machine carries out the steps.
The machine proposes. A person with standing decides.
The machine chooses and acts inside limits a person drew.
"Whatever has not been expressed has not been delegated."The Constitution of the Frontier, the Intent Contract
The machine operates. People audit the outcomes afterward.
"A system you cannot turn off is not your infrastructure. You are its."The Constitution of the Frontier, Principle 13
The system proposes where the limits should sit... but governing remains human.
"Drafting a suggested limit and authorizing that limit are different decisions."Decision-Scoped Delegation for AI Systems, v1.1
Rungs 1 to 4 build the evidence. A proposal at Rung 5 travels back to Rung 0, where a named human decides again. Trust extends on evidence; verification stays human.
"Authority moves by decision, not by drift."The Constitution of the Frontier, Principle 6
Pick a decision and a rung. You'll get the six parts of a written grant, tuned to the rung you chose. This is a thinking tool: nothing is saved or sent.
The path runs from noticing what you're handing over to proving who answers for it. Each stage ends with something in writing, so progress is visible to a board, a regulator or a class.
What are we actually handing over?
You keepAn inventory of consequential actions and a census of the agents you run.
Which rung is this on, and who put it there?
You keepA register, a grant for each class of decision, an Agent Personnel File and a named sponsor for each system.
Can our people run the rules they adopted?
You keepA front door for new AI requests, instruction files, a safe way to report problems, and a weekly, monthly and quarterly rhythm.
What can the agent make happen, and what stops it?
You keepRule-to-Control Matrix rows, a working gate on one real action, and Prove It test results.
Where did the saved time go, with evidence?
You keepA Dual Ledger, an options memo, a day-90 readout and the cost per accepted outcome.
When this goes wrong, who apologizes?
You keepA ratchet-down log, a six-month review and an operating agreement for each agent.
And it loops. Every incident, model change or departed sponsor sends authority back to stage 2, to be decided again.
A narrated walk from the drift problem, up every rung, around the Trust but Verify loop and across the six-stage journey. Captions are built in.
Two one-page briefs to share with your team, the technical white paper behind the method, and a short podcast episode that puts the ideas in context.
The core thesis on one page: why it matters, the six rungs as a loop, the three questions and how organizations start.
Download PDF
The engagement model: four steps to begin, three engagements on one journey, and your first six weeks.
Download PDFThe research specification behind the ladder: delegated authority scoped to one decision at a time, with a runnable reference model. Grounded in The Constitution of the Frontier. Version 1.1, October 2026; not yet peer reviewed.
DOI 10.5281/zenodo.23126767 Read on ZenodoWhy a shutdown switch isn't enough, why approving what you can't evaluate is ceremony rather than governance, and why capability should stay separate from authority. Stanislav Petrov's 1983 decision makes the case.
Listen to the episodeWhether you want to carry the method into your own field or bring it into your organization, there's a clear first step.
For change leaders, consultants, educators and AI leads who want to apply the Delegation Ladder as an AI-centric change management method.
For organizations
A menu of professional services across the whole journey, starting with Phase 1: the six-week AI Governance and Enablement Foundation.
