Guide

The Four Levels of AI

From a chatbot you ask questions to agents that run recurring work on their own. The four levels, where my own team sits, and how to move up one.

At a glance: Time: About 10 minutes; You need: A list of what AI does for you now; You get: Your level, and the next step up

This week I sorted my own business into the four levels of AI. My team sits at Level 3.

Level 3 means agents do the work, and I kick each one off. Level 4 is when they run on their own.

Almost everyone I talk to is still on Level 1. They paste something in, get an answer, and start over the next day.

Sources checked October 5, 2026. Where my team sits is my own read of my own setup, October 5, 2026.

The four levels

Diagram: Level 1: Answers, then Level 2: Actions, then Level 3: Delegation, then Level 4: Loops.
  • Level 1 is useful for thinking and drafting. The work still waits for you.

  • Level 2 gives it hands: your files, your tools, one job at a time.

  • Level 3 is a small team of agents, each with its own job.

  • Level 4 runs until the job is done, or asks a person for help. My LEVELS scorecard, August 8, 2026

Your level is what you can prove

Your level is the highest one you can prove. Not the highest one you tried.
  • Level 1: you use AI to brainstorm, research or draft.

  • Level 2: it takes one approved action in a real tool.

  • Level 3: one request gets split across agents with the right access.

  • Level 4: something starts it, it checks the result, and it retries or asks for help.

Where my team sits: Level 3

Diagram: I kick it off, then Agents split the work, then I say yes.

My content team is agents. One finds ideas, one writes, one reads the numbers. Each run still starts with me.

That's the gap to Level 4. The work is done by agents, but the start is still my hand.

Move up one level

The steps: 1. Level 1 to 2.; 2. Level 2 to 3.; 3. Level 3 to 4.

Anthropic, who make Claude, say the same: start simple, and add more agents only when simpler steps fall short. Anthropic: Building effective agents

Copy and paste: a Level 2 start

The jump from 1 to 2 is smaller than people think. Give it something of yours, then ask it to do the task, not describe it.

Here are my notes from this week:
[paste them]

Draft the customer update from these notes.
Use only what's in the notes.
Mark anything you're unsure about with [CHECK].

Keep a person at the gate

Skip: Automating a job you can't define yet. Do: Define what done looks like first.; Skip: Letting it spend or post on its own. Do: A person signs off on money and anything public.; Skip: Jumping from Level 1 to Level 4. Do: One level at a time.

Anthropic says agents should pause for human feedback at checkpoints. That's where my yes sits. Anthropic: Building effective agents

Sources

  1. SAMethod: The four levels of AI for a service business (LEVELS scorecard, August 8, 2026)

  2. Anthropic: Building effective agents

Connect

Not sure which level fits your business? My free Revenue Opportunity Review looks at where AI would actually pay off. Request a free review

samethod.ai · Case studies · sameer@samethod.ai

Follow along: Instagram · YouTube · LinkedIn · X · Threads

Keep a copy

Download the PDF edition. Check the article’s date note for updates.

Download guide

Get it by email

I’ll email you the link so it’s there when you need it. No weekly emails.

Want me to look at your business? Get a free review →