I Started With Basic Prompts. Now I Have AI Agents Doing the Work While I Sleep.

A while back I wrote about how a small taste of AI at work turned into something much bigger than I expected.

This is the next chapter of that, and honestly it still feels a bit mad to type.

I now have AI doing work for me when I'm not there. Not "I asked it a question and it answered." Actually going off, on its own schedule, doing a job, and having it ready for me when I come back.

Let me tell you what that actually looks like, because when I was starting out I would have found this completely mystifying.

Where I was

Not that long ago, my AI skills were: open the chat, type a question, get an answer, copy the answer out.

That's it. That's genuinely where everyone starts, and there is nothing wrong with it.

The problem is it doesn't scale. Every single thing I wanted had to be asked for, every single time. I was the bottleneck. If I didn't sit down and prompt it, nothing happened.

What I've actually built

Three things changed that, and none of them required me to become a developer.

Content that gets drafted without me. I set up a saved skill, basically a written set of instructions AI follows every time that knows my brand: my colours, my fonts, my sizing, the way I write titles and captions. Then I attached it to a schedule. Now drafts get made on a regular basis, pulled from a running list of topics I top up whenever an idea hits me. I'm not starting from a blank page anymore. I'm reviewing, choosing and publishing.

AI that goes and looks things up for me. I connected AI to my browser. That means it can open my social media accounts, get into the analytics, read my numbers, and pull out what's actually happening. Reach, saves, clicks, which posts are landing, which topics are carrying the account. I used to click through all of that myself and try to hold it in my head.

A dashboard that rebuilds itself. All those numbers now land in a proper visual dashboard I can open any time. And it refreshes on a schedule — every Wednesday morning, before I'm properly awake, the numbers get pulled and the whole thing gets rebuilt with fresh data and a short summary of what changed.

I didn't code any of that. I described what I wanted, clearly, and refined it when it wasn't right.

Hang on…. is this an agent?

I actually stopped and asked this, because "AI agent" is one of those phrases that gets thrown around constantly and rarely explained.

The answer is yes, and here's the simplest way I've found to think about it.

A tool does one step when you press the button. You ask, it answers. You're driving.

An agent goes after a goal across lots of steps, and adapts when things don't go to plan. Nobody's driving.

My Wednesday task wakes up on its own. It works out how to get into my analytics. It reads a page it hasn't seen before. It decides which numbers matter this week. It notices when something's shifted and tells me. If a page looks different than expected, it works around it.

That's the difference. It's not following a rigid script I wrote, it's pursuing an outcome I described.

And the honest bit: that's a genuinely different relationship with AI than the one I had six months ago. I've gone from asking to delegating.

There's a next level, and I've started climbing it

Here's the thing I've realised, and it's the whole reason I'm still going.

There's a ceiling on the way I'm doing all this right now. It works, I use it every single week, but it's a clever workaround rather than a proper connection. It's slower than it needs to be, and it's more fragile than I'd like.

There is a proper version. And it isn't a bigger prompt or a better tool. It's a different kind of skill entirely, and a much harder one.

I'm not going to explain it yet, because I've only just started learning it and I'd rather build the thing before I write about it.

What I will say is this: it's the piece that takes all of this from "clever personal setup" to something businesses would genuinely pay for. Because most of the slow, messy, repetitive processes inside real companies live in software that nobody has connected to AI yet. Anyone can plug AI into the popular apps. Very few people can reach the awkward stuff.

That's where I'm heading. I'll show you when I get there.

If you're just starting

I want to be straight with you about the path here, because from the outside it can look like a leap.

It wasn't. It was: learn to prompt properly. Then save the good instructions so you're not rewriting them. Then attach those instructions to a schedule so they run without you. Then connect it to something real.

Each step only made sense because I'd done the one before it. None of them required me to be technical. All of them required me to keep going slightly further than felt comfortable.

You do not need to understand agents to start. You need to start, and the agents will make sense later.

I'm still very much in the middle of this. But I've got AI working while I sleep now, and a year ago I didn't know what a prompt was.

Follow the spark. Then automate it.

Welcome to your future.

FutureAIGirl