How much leash? Levels of autonomy
Let it auto-run everything, or approve each step? Get this wrong and you either babysit or get burned.
The idea inside
Autonomy is a dial from suggest to teammate; match it to how well you can verify it.
After this lesson
You can match an agent's autonomy level to a task's clarity and how easily you can verify it.
Where it leads
And when one agent isn't enough, you reach for several.
Inside this lesson
That's the real lesson stage, paused. Claim your pass to operate it.
See how AI actually works, end to end.
This lesson is one stop on the full arc. Unlock all of it, and keep it for life.
What you get
- The 34-lesson main path, a finishable route from a word to agents
- Goal tracks for using AI at work and building AI features
- Boss labs that make you apply a whole act, not just recognize it
- Spaced recall that brings each idea back before you forget
- Course memory: every term defined, with links to where it first appears
- A shareable capability card when you finish the main path
- Lifetime access on every device, every future lesson included
Not videos to watch. You predict, operate the machine, then prove it. That is why it stays.
99 interactive lessons and challenges. No videos, no code.
Free launch pass: lifetime access, no card needed
New here? The first lessons are free to try. Start with lesson 0.1
What this lesson shows
Autonomy is a dial from suggest to teammate; match it to how well you can verify it.
The question it opens with
Let it auto-run everything, or approve each step? Get this wrong and you either babysit or get burned.
The walkthrough, in the lesson's own words
- Drag from 'you drive' to 'it drives'. Watch the cost flip.
- Now pick the task. The right setting moves with it.
- Autonomy is a staffing call: more leash where it's scoped, checkable, and undoable.
- Same dial, opposite costs: low and you do the work, high and you find out later. Nothing here is about how smart the model is.
- The leash depends on undo-ability as much as verifiability: auto-accepting edits under version control is not the same as sending a refund email.
- The four detents, suggest → approve each step → hand off → autonomous teammate, run from "you drive" to "it drives". Higher isn't better.
- Scoped + checkable + undoable earns a long leash; fuzzy, hard-to-verify, or irreversible keeps you close. It's a staffing decision, made per task.
- Suggest is chat: it drafts, you copy and apply everything yourself.
- Approve each step is the default ask mode in agent tools: every edit or command waits for your ok.
- Hand off is auto-accept mode: the agent runs the whole task and you review the finished result.
- Autonomous teammate is a background agent: it works on its own and you review at the end, like a pull request.
- You give an agent a vague goal and walk away; it does the wrong thing confidently. What dial was set wrong?
- Autonomy: how much leash the agent has before it must check in. A fuzzy, hard-to-verify goal can't safely absorb a long leash, so lower it to approve-each-step. Save high autonomy for tasks that are scoped and checkable.
- This task sits in the checkable corner, but there's no undo once it ships, so it stays keep-close. Reversibility is the third axis this map can't show.
Key takeaway
You can set autonomy per task, like a staffing decision, more leash where it's scoped and checkable.
What you can do after this lesson
You can match an agent's autonomy level to a task's clarity and how easily you can verify it.
Check yourself: How should you set an agent's autonomy level?
- Match it to how scoped the task is and how easily you can verify it(correct)
- Always maximum autonomy
- Always minimum autonomy
- Pick at random
Give more autonomy when the task is well-scoped and easy to check, less when it is open-ended or hard to verify. Match the leash to the risk.
Where it leads: And when one agent isn't enough, you reach for several.
This is the written summary. The lesson itself is interactive: you predict, drag and operate the mechanism above, and the reveal answers you.