Orient / Assist / Supervise / Delegate

The Trust Ladder

The trust ladder sorts AI work by one question: how much are you still checking? There are four rungs, Orient, Assist, Supervise, and Delegate, and you move up only when you can defend the decision to someone who asks.

Most advice about AI sorts tools by what they can do. That is the wrong axis for a business. What matters is how much of your judgment the work still passes through, because that is what decides your exposure when something goes wrong.

Every course on this site is labeled with one of the four rungs. The label tells you what the course expects of you and what it hands over. You can stop at any rung and stay there. Most business use of these tools should live at Supervise, and plenty of businesses do fine at Assist forever.

Nothing about the ladder is automatic. Moving up is a decision you make, in a specific business, about a specific job.

01
Rung 1 of 4

Orient

Understand the tools before you hand them anything.
Orient is the learning rung. You are figuring out what these tools actually are, where they are strong, where they fail, and what the words mean when a vendor uses them. You are not putting AI into a workflow yet.
What you do at this level
You try AI on small, low-stakes things and read every result closely. You compare what it produced against what you would have written. You learn the difference between a tool that writes text and a tool that can take actions in your systems, because that difference decides everything about risk.
Before you move up, you should be able to answer
  • What is the difference between a chat assistant and an agent?
  • What can this tool reach, and what can it change?
  • Where does a person have to approve something before it happens?
  • Which jobs in my business should stay with a person no matter what?
Courses at this level

What Is an AI Agent

View course →
02
Rung 2 of 4

Assist

You do the work. The AI helps. You review everything.
Assist is where AI enters real work. You write the ask, the tool produces a draft, and you decide what ships. Nothing leaves your desk in your name without your eyes on it.
What you do at this level
You brief the tool properly instead of asking it a vague question, using H.E.L.P. or a method like it. You review every output before it is published, sent, or filed. You start keeping the prompts that worked, so the second month costs less effort than the first.
Before you move up, you should be able to answer
  • What did I tell it, and what did it fill in on its own?
  • Would I catch a confident, invented detail in this draft?
  • What does good look like for this task, written down before the work starts?
  • Which recurring deliverable am I now producing faster, and can I prove it?
Courses at this level

The H.E.L.P. Framework

View course →

AI for Marketing

View course →

Fundamentals of SEO with AI

View course →
03
Rung 3 of 4

Supervise

The tool does the work while you watch.
Supervise is where the tool runs a whole task instead of producing one draft. It works across your files or your connected tools, takes several steps, and hands back finished work. Your job stops being production and becomes review.
What you do at this level
You scope a task small enough to check and reversible enough to undo. You watch it run and stop it when it drifts. You read permission prompts instead of clicking through them. When it finishes, you check the result against the source material, not against the tool’s own summary of what it did.
Before you move up, you should be able to answer
  • Can I undo everything this task touched?
  • Where exactly did this number or this quote come from?
  • At what point should this have stopped and asked me?
  • Has this same task run cleanly enough times that I would let it run unwatched?
Courses at this level

Intro to Claude Cowork

View course →

Intro to Claude Code

View course →
04
Rung 4 of 4

Delegate

You set the rules. The work runs inside them. You check the record.
Delegate is governance. Recurring work runs without you starting it each time, inside limits you wrote down. You are no longer watching each run. You are responsible for the boundaries and for reading the trail afterward.
What you do at this level
You put rules where they are actually enforced instead of where they are merely suggested. You give each job the narrowest access it needs. You keep a log that records what ran and what it changed, and you read that log on a schedule. You hand the whole configuration to a colleague, so the person covering for you in August runs exactly what you run.
Before you move up, or before you stay here honestly, you should be able to answer
  • If this misbehaves at two in the morning, what is the worst it can do?
  • Which of my rules would survive the tool deciding to ignore them?
  • Who reads the audit trail, and how often?
  • Has this job outgrown the tool, and should it be real software instead?
Courses at this level

The Four Surfaces of Claude Code

View course →

Intro to the Codex App

View course →
Coming soon

The Agent Seat

A hands-on Delegate capstone on deploying and governing a supervised agent. Not open yet. Email info@nantucketai.com to hear when it is.

Where should I start?

Start at Orient unless you already use AI in your work most days. That is one course, about two and a half hours, and it is the one that makes every other course make sense. If you already use a chat assistant daily and you are past the “what even is this” stage, start at Assist with The H.E.L.P. Framework and come back to Orient when somebody tries to sell you an agent.

If you are not sure, take the Orient course. It is free, it is short, and nothing later on assumes you skipped it.

Start the free foundations course
Browse all courses →