# Autonomy levels · grade the task, not the AI

One question grades every task: **what happens if this is wrong?**
Then two follow-ups: who notices, and when? Can it be undone?

**Level 1 · Fully autonomous.** Wrong costs a minute.
Research a company. Draft the first message. Tag the inbox. Score a lead. It runs; you read a report.

**Level 2 · Human-assisted.** Wrong is a bad draft someone would have caught.
Proposals, client replies, the week's posts, call summaries. The AI drafts; a person edits and sends.

**Level 3 · Human-led.** Wrong costs money or reputation.
Invoices, hires, the first touch to a lead, anything that leaves the building. The AI prepares, stops, and a human presses send.

**Moving between levels.** A task earns a level up with a clean record. It loses a level the day it's wrong.
The grade is on the task, never on "the AI".

## How to use it
1. List every recurring task in your business. (Mine came to 35: 15 autonomous, 13 assisted, 7 human-led.)
2. Write the level next to each one.
3. Build the Level 1 tasks first. Most hours back, least risk.

| Task | What happens if it's wrong? | Level |
|---|---|---|
| | | |
| | | |
| | | |

---
The Paper-First Method by Sohaib Waheed · paperfirstmethod.com · free to use with credit (CC BY 4.0)