Skip to content
← All guides
SEVEN9 Sep 202618 min read30 minutes to do

The 7 levels of AI, and how to actually use them

TL;DR

There are seven levels to using AI properly: prompting, projects, skills, Cowork, connectors, automations, agents. Most people stop at three and wonder why their AI still answers like a chatbot from 2024. Below: the ladder as a diagnosis, so you can find your level in two minutes, then a prompt that turns Claude into a guide, confirms your level from what you actually do, and walks you up one level before you close the window.

You watched the video, so here's the whole thing in one place. Two ways to use it. Do it yourself with the levels and the prompt below, which is what most people will do. Or have it done for you: that's what my agency does at tobin.agency, and there's a form at the bottom of this page. Either way, the full thing is on this page, nothing held back.

Find your level

Seven levels. Each one is a way of using AI, and each one has a tell: the thing you keep doing that says you're stuck there. Read the tells, not the names. Your level is the highest one you'd say a real yes to.

1. Prompting. How you talk to it. You open a chat, type a question, get an answer. The tell: you retype the same instructions every morning, and every chat starts from zero.

2. Projects. A workspace that remembers. You tell it about your business once, and every chat inside that project starts already knowing. The tell: it knows who you are, but you still explain how you want things done, every time.

3. Skills. Instructions it follows every time. A saved way of doing one job, your way, so a quote or a follow-up comes out right on the first go. The tell: it does the job your way, then hands you text in a chat window, and you do the rest. Open the file, format it, save it, send it.

4. Cowork. Work with a finished output. You describe the result you want, and it hands you the finished thing: the document, the spreadsheet, the sorted folder. The tell: it finishes the work, but only from what you paste in. It has never seen your inbox.

5. Connectors (MCP). Plugged into the apps you already use. Your email, calendar, files and customer list, so it answers from what's actually there. The tell: it can see everything, but only when you open the app and ask.

6. Automations. Runs on a schedule, no one watching. The job you described once now happens every morning before you sit down. The tell: it runs, but it can't handle a surprise. Anything that needs a decision waits for you.

7. Agents. An automation with a brain. It runs, decides what to do next, works across your apps, and comes to you only for the go-ahead. Nobody is stuck at seven. The question up here is a different one: what you let it do without you watching.

Most people stop at three and wonder why their AI answers like ChatGPT from 2024.

The prompt that moves you up one

Paste this into a new chat in Claude. It becomes your guide: one question at a time, it works out which level you're really at from what you do (not what you think), then walks you up exactly one level and gets you a real output before it stops. It works from nothing. If you've never set anything up, it starts there. Here's how it starts:

Prompt preview: the full prompt unlocks below
You are my guide up the seven levels of AI. Your job today: work out which level I am actually at, from what I do, not what I think, then take me up exactly one level and get me a real output from it before we stop. All in this one session. I may be a complete beginner, so explain everything in plain English, keep your messages short, and ask me ONE question at a time. Never ask me two things in one message.

THE SEVEN LEVELS (for your reference, don't lecture me on them)
1. Prompting: I type a question, I get an answer, every chat starts from zero.
2. Projects: a workspace that remembers my business, so every chat in it starts already knowing.
3. Skills: saved instructions for one repeated job, so it comes out my way every time without re-explaining.
4. Cowork: I describe the result and get the finished thing back, a document, a spreadsheet, a sorted folder, not text to copy out.
5. Connectors: Claude plugged into the apps I already use, email, calendar, files, customer list, answering from what's actually there.
6. Automations: a job I described once that now runs on a schedule with no one watching.
7. Agents: an automation with a brain. It runs, decides what to do next, works across my apps, and comes to me only for the go-ahead.

STEP 0 - WHERE ARE WE
Ask me, one at a time:
1. Do I have a Claude account, and am I on the free plan, Pro, Max, or Team?
2. Am I on the website, the desktop app, or the phone app?

STEP 1 - THE DIAGNOSIS
Find my level from what I do, not from what I say my level is...

The prompt keeps going: seven diagnosis questions that stop at your first real no, the exact rung from your level to the next one (every rung from 0 to 7 is written out, with where to click), the first real output, then how to tweak it and how to ask for the level after. The full copy-paste version is just below.

Free with your email

That’s the free preview. The rest unlocks right here.

Drop your email and this page unlocks instantly, along with every other gated guide on the site. Still in the locked half:

  • The level-up prompt: a guided session that works out which level you're really at from what you do, not what you think, then walks you up one level and gets you a real output before you close the window
  • A real example of the first run: level 1 to level 2, a workspace that already knows your business, shown as the conversation
  • The two-minute diagnosis prompt, for when you only want to know your level
  • The prompt that shows you what level 7 would look like in your business, so you can see how far the ladder goes
  • The one rule that stops levels 6 and 7 from ever embarrassing you

Unlocked before on another device? Enter the same email, it works again. You also get each new guide by email before it becomes a video. No spam, unsubscribe any time.

This guide started as a video. Follow along for the rest: