5 steps to your first ai employee


There's a whole genre of AI videos I can't stand…

The ones where the creator tells you to set up an army of 50 sub agents, drops a GitHub repo underneath it, and calls it a starter kit.

I hate them because they always end the same way...

  • You open it
  • You can't work out what any of the 50 agents do
  • You end up running none

Or you run them all, and it cripples your performance and skyrockets your costs

It might feel really tempting to use all 50 agents. It might feel like you all of a sudden are way more productive. You’re not though. You just slowed down your process greatly.

Seeing this a lot more now too with Grok Bot coming out, and people just immediately make 20 agents.

Which brings me to the part those videos skip:

You actually don't need 50 agents. You just start with 1 you build yourself.

And yes, I know that trying to invent a job for an AI agent is harder than downloading somebody else's repo.

But you don't have to invent anything.

Trust me, you already know what your AI agent should do.

Because you did it yourself yesterday, by hand, on your own computer.

You just need to find that job and get your agent running it.

So let me show you how to do it in 5 steps.

1. Brain dump everything about yourself into your agent

Before you ask it for anything, tell it:

  • What you do
  • What you're building
  • What you're paid for
  • And what you're trying to get done

I’ve told my AI agent that I run an AI YouTube channel, own a SaaS called Creator Buddy that does over $300,000 in ARR, just raised pre-seed for Henry Intelligent Machines, put out a newsletter twice a week, and so on.

You need to do this step because your agent can't help you with delegating a task if it doesn't know what your job is.

So don’t rush this, unless you want everything to come back generic.

2. Carry a notepad for a day and write all manual tasks you do on your computer

Yes… paper, not an app.

And it should sit next to you all day.

That way you catch tasks as they happen instead of trying to remember them at the end of your workday.

So as you work, write down what you do.

Here are a few examples:

  • Writing content
  • Writing emails
  • Managing your calendar
  • Answering people in Slack

Don't shortcut this one.

The jobs worth delegating are usually the ones you'd never have thought to write down.

Now, a piece of paper covered with tasks doesn't do anything on its own…

3. Ask your agent what it can take off your plate

At the end of the day, type every task from that paper into your agent (or take a screenshot and paste it in).

And ask it what it can take off your plate.

That way you never have to think up workflows yourself.

Here's a prompt I use with the list:

Based on everything you know about me, my goals, my jobs and tasks, and what I do on a daily basis, what are some workflows and use cases we can implement?

So as you can see, I'm not the one deciding what gets taken off my plate.

My agent works that out from everything it knows about me.

And some of the jobs the agent comes back with, I’d never have thought to hand over in the first place.

4. Choose 1 task (don't overthink it)

Not the most impressive job on the list. Not 3 of them. Just 1.

This is the step the 50-sub-agent repo skips, and it's the only one that decides whether you'll still be running any of the agents next week.

And if you're worried about picking the wrong one, remember you can always run the prompt again and pick another.

So pick 1 task and watch your agent work.

5. Add a second job only once you've stopped checking on the first

When you do stop babysitting the first one, run the same prompt from step 3 and take the next job off the list.

From there, once you feel like you have your agent fully occupied, you can think about expanding to 2nd and 3rd agents with their own distinct jobs.

Btw, you don't have to figure out what agent to build (and how) on your own.

Every Friday I run a live session inside the Vibe Coding Academy where 800+ members are shipping their own apps. Just bring your list, and we'll pick your first job and get it running together.

Join the Vibe Coding Academy here.

Keep shipping,

Alex

Ship/It Weekly

The weekly newsletter for founders and builders using AI to ship faster—practical tactics, tools, and real-world lessons from Alex Finn.

Read more from Ship/It Weekly

6.44 billion people have never used AI. Yes... not once. Even though the most powerful technology ever created is free to try, most people have never touched it. Which is why everyone using AI to build and automate their lives has an OBSCENE advantage right now. To help you stay on the cutting edge, I've put out 3 posts over the last couple of weeks. 1. 77.6% of Earth has never used AI In a decade 75% of this chart will be red. Until then you have the greatest arbitrage opportunity there is....

I have never designed anything in my life. I'm very unartistic, and I just stink at all of that. So when I started using Claude Design, I experimented a ton just to figure out what I even liked… I asked questions I tried things 10 different times And I hit my limits fast without creating anything good Ever had a similar situation? Chances are, you made the exact same mistake I did… You opened the best AI tool before you knew what you wanted. But, as you probably already know, every AI tool...

Every AI agent you run lives in its own little box… Your OpenClaw can't see the context from Hermes Your Hermes can't see your conversations with Claude Code And your Claude Code can't see the work you've been doing in Codex But Jack Dorsey just released a new multi-agent chat app that finally lets them read each other's work. It's called Buzz. And there's so much more to it that no one else is talking about. Which is why I put together a full breakdown of Buzz in a YouTube video 2 weeks ago....