How I Use AI: Stop Prompting and Start Hiring

Before we get started…

If AI still feels like a toy…fun to play with, but not actually increasing output or ROI (more on that in a minute)…

…it's a sign the real work in your business is still running through a handful of humans (including you). And when they're out, everything stops.

The solution? Clone them.

This Wednesday (September 23rd at 1 pm CT), my partner, Richard Lindner, and I are running a live 90-minute workshop called The New AI Org Chart.

We'll capture how your best people do their work, and "clone them" into an AI employee that never calls in sick, never takes a vacation, and never asks for a raise.

(And no, you don't need to get rid of the human you just cloned… we'll also talk about why that's a super-stupid idea.)

Click here to save your seat.

Ok, on to this week's article…

That was me this time last year.

I was spending a fortune on all the latest and greatest AI tools for me and my team, and all we got was a slow, kinda crappy editing assistant.

Every session started from zero.

I'd explain the business, explain the customer, explain what we sell and why anyone buys it…and then get back something that sounded like every other company in our category.

So I'd rewrite it. Every time.

But everything changed when I stopped trying to write better prompts, and I started treating AI like an employee I'd hired.

Today my AI "hires" turn out real, finished work across every department in my business.

Now, I'm going to show you the 5-step process (and the prompts I use) that finally cracked the AI code for us.

Let's get into it…

Two years ago I wanted to get serious about my personal brand.

And to do that, I needed Matt, my Head of Marketing.

Matt's a pro. He knows his stuff. He also knows everything there is to know about our company and how my personal brand can be leveraged to grow it.

Unfortunately, Matt is one of the most valuable (and busiest) people in our company…

…which is exactly why I couldn't justify taking 10-20 hours of his time to help me "figure out LinkedIn."

So instead of booking time I couldn't get, I built a Project and "hired" a mini-Matt. (I use Claude, but ChatGPT and Gemini have the same thing under a different name.)

What made Matt so valuable was all the knowledge, skills, and context he had before I ever asked him a question.

A brand-new chat window knows none of that, but a Project can, and that's the whole difference between a task and a role:

That's why I name all my projects after the ROLE they'll be doing for me.

Here's a peek at my AI employee roster (and this is just the ones that fit on one screen)…

Strategist. Consultant. Copywriter. Content Producer. Newsletter Writer (yes, this issue started there). Social Marketer. YouTube Strategist. Brand Strategist. Instructional Designer. Book Editor.

Every one of them performs an essential role that I need on a daily or weekly basis, and every one of them knows more about me and my business today than it did a month ago.

One more thing on this…

Don't build one genius employee to rule them all. AI can only hold so much in its "head" at once, which is why a Project that tries to do six jobs will do all six of them worse than six Projects doing one job each.

You wouldn't hire a Head of Marketing by saying "there's your laptop, good luck."

The same thing applies here.

Every Project has an "Instructions" field, and instructions are just a job description for your AI employee. What this role does, what it's responsible for, what experience it's supposed to already have.

Here's the prompt I use to write them:

Tweak the output so it matches the needs and nuances of your business, then paste it into your Project's instructions box.

This process takes less than 10 minutes, and it's the difference between a generic assistant and a true coworker.

This is the step everybody skips, but it's the one that really moves the needle…

AI is pretty smart. The problem is, it doesn't know anything about you or your business.

Ask it for a marketing plan, and you'll get a generic plan that's the average of every marketing plan that's ever been published anywhere, which then gets blended into a plan that fits nobody.

So before you ask for a single deliverable, feed it the same things you'd give a new hire in week one:

Here's the uncomfortable part.

Most business owners don't have any of this written down anywhere. It lives in your head, which is why the business can't run without you and why the AI can't either.

Assembling these assets fixes both.

Once a new hire gets onboarded (Step 3), they then get additional training on their craft.

The same needs to be true for your AI employee…

Now is when you want to "feed it" all the best courses and trainings, including the ones you've already invested in and haven't consumed.

Just think about all the things you've paid for and never finished:

All of it goes in the Project, either uploaded as a "File" or just copied and pasted into a chat.

Unlike your human employees, the AI will actually consume it and remember it.

When a Project is fully tested and delivering quality output consistently, that's when it's time to turn it into a Skill.

Projects are great, but they're still largely limited to a single account. But a Skill is a transferable file you can hand to your assistant, your marketer, your agency, or even another Project…

…and it runs the same way for them as it does for you.

There are two ways to build Skills in Claude, and neither requires you to write a single line of instructions yourself:

METHOD 1: Use the built-in Skill Creator. Open the chat where the Project just nailed the task, type "/skill-creator," and tell it to turn that chat into a skill. It reads what you asked for, what came back, and what you corrected along the way, then writes the skill for you.

METHOD 2: Use the "Record a Skill" feature. Click the "+" button, choose "Record a skill," and just do the task the way you normally would. Claude watches, then writes up the process as a skill your team can run without you.

Either way, keep feeding it new data, examples, and best practices.

Every time something comes back great, save the example to the Skill. One or two pieces of your best output teach more than any set of instructions you could write.

That's the AI Employee loop:

Hire the role → write the description → onboard them into your business → send them to training → promote the good ones (to Skills).

⚡️ Action Step: Pick the one task you delegate to AI most often and turn it into a Project this week. Give it a name, write the job description, and upload three documents about your business before you ask it for anything. Then run the same request you ran last week and compare the two.

Give it a shot and let me know how it goes…

-Ryan

Ryan Deiss
Co-Founder and CEO, The Scalable Company

P.S. Everything above is how I got AI to stop being a toy and start producing work I actually ship.

But if you really want to take this concept to the next level, you need to register for our upcoming workshop: The New AI Org Chart.

In 90 minutes we'll show you how to "clone" your best people (their decisions, their standards, their context) and install their knowledge into an AI employee that keeps executing even when they're not around.

And this workshop isn't just about learning…

…you'll actually leave with one working AI employee and a system to build the rest.

Click here to save your seat.