I got my hair cut this week.

Nothing unusual there. Same barbershop I've been going to for years. Same bloke, same chair, same conversation about how my day at work has been.

Except this time, I overheard him explaining something to the other barber in Arabic with more enthusiasm than I've ever heard from him before.

Bear in mind I don't speak Arabic, but luckily ChatGPT seems to be pronounced the same universally.

So I asked him, and he was telling me how he'd connected ChatGPT to a Google Sheet.

Now he can record his working hours just by telling ChatGPT what he'd done that day.

And he was absolutely buzzing about it.

The whole time I was sitting there quietly smiling, because I understood exactly how he felt.

I remember when I first got my hands on the early version of ChatGPT.

It felt like magic.

Especially at university. I could paste in some random assignment brief and have a large chunk of the work back in minutes.

Which, on reflection, probably tells you plenty about my degrees.

But that was only the beginning, because I couldn't leave it alone after that.

By the end of uni I'd built myself a full ecosystem of prompts, each one designed to squeeze the highest marks possible out of any type of assignment.

It stopped being about getting help and started being about seeing how far I could push AI.

As I started taking on bigger projects, the tasks I gave it became more complicated, and the technology kept getting better at handling them.

And I loved it. Every part of it.

Except the further I pushed it, the more I started noticing a rather strange problem.

THE PARADOX

The more capable AI becomes, the less of the actual work I seem to do.

And weirdly, the less satisfying some of it feels.

I'm building more than ever and I'm less involved than ever.

Those two things are happening at the same time and I can't decide how I feel about it.

To be clear, I'm not anti-AI. Quite the opposite. AI has let me build things I couldn't have built otherwise, and I'm not going back to doing everything by hand.

But I've started to wonder what my future job becomes when most of the execution can be handed off to a machine.

HOW I ENDED UP PRESSING BUTTONS

If you've read the previous weeks, you know I'm stuck in a cycle of designing the Clickrabbit site and rebuilding small bits of Life of Jake.

Most of that gets done through something called Agentic AI.

For anyone who doesn't speak fluent nerd, here's the plain version.

ChatGPT is a box you type into and it types back.

Agentic AI is the next step up. It doesn't just suggest code, it goes into your project and does the work. It opens your files, writes the code, fixes the bits that are wrong, runs the tests, and keeps going until the feature works.

Which is truly insane, because I can't write code properly.

Hand me a complicated bit of code and I can roughly guess what it's trying to do.

Ask me to fix it and we're getting nowhere.

But the beauty of this is that all of us can describe what we want.

It's like renovating a house. You might not know how to lay a brick or fit a kitchen, but you can still walk in and explain exactly what you want changed.

Describe a complicated website component well enough, and the agent builds it.

Something that would've taken me weeks, gets done in an evening.

Here's the problem though.

I'll usually have two of these agents running at once, each on its own task, so both jobs get done in about an hour.

Which means for that hour, I've got nothing to do but wait.

You'd think that's the dream. Kick back, watch Netflix, get on with your own thing while the work does itself.

Except they don't work unsupervised. Every few minutes, one of them stops and asks for permission.

"Would you like to execute this command?"

"Can I utilise this tool for that task?"

The usual solution is to give the AI full access to your computer and let it crack on. But we've all seen enough movies to know how that one ends.

So the sensible option is to approve things yourself.

Here's the flaw. If you don't write code, you've got no real idea what you're approving.

So you look at it, quickly decide it probably won't blow up your laptop, and grant the permission.

Which is insanely unsafe, because you're waving through technical decisions you can't assess. The alternative is sitting there asking questions you don't have the knowledge to ask.

And this is the part that caught me out.

Agentic AI has quietly taken me out of my own work. I used to be the one doing it, and the AI was just the thing handing me a helping hand when I got stuck. Now it does the whole job and I'm the bloke standing behind it saying yes.

Which leaves an obvious question. How are you supposed to approve something you don't understand?

You can't. You guess, you press enter, and you carry on. And once you've done that enough times, you've got no real idea what's going on inside your own product.

So most evenings, when I'm supposedly working, I'm just pressing Enter, playing Clash Royale, and glancing up every few minutes to make sure the machines haven't gone rogue.

SATURDAY

Then last Saturday I was out with my mates, going for a spin in the car.

Before leaving the house I'd set the agents a large task, and all evening I kept checking my phone to approve the permissions they wanted so they'd keep working.

Technically, that's incredible. Work was getting done while I was out with my friends.

Except I kept checking my phone. And half the time I didn't understand what I was approving.

I wasn't working, but I wasn't switched off either.

If I have to keep checking my phone to press a button I don't understand, then I'm not free.

I'm just outside.

HAS THE ART OF BUILDING CHANGED?

Here's where it gets interesting, because I love building things.

I've been messing with websites, designs, and little systems for years. There's something deeply satisfying about having an idea, struggling to get it working, and eventually watching it come to life.

It's the thing I'd do anyway, even if nobody paid me.

AI has changed that process completely.

Before, the challenge was figuring out how to make the thing. Now the challenge is figuring out what you want to make.

Take Clickrabbit. I'm building a website with custom animations, interactive bits, and functionality I wouldn't have had the technical ability to create myself.

The agent writes it. I still decide what belongs on the page, how everything should look, and whether it actually communicates what the business does.

The weird part is that building has started to mean something different to me.

I'm no longer the person putting every piece together. I'm becoming the person who imagines the finished thing and directs how it gets built.

Which raises an uncomfortable question.

If the machine is doing the building, am I still a builder?

I think I am. But the answer probably depends on how involved I choose to be.

THE STRANGE ART OF DOING NOTHING

I've spent years learning how to do things.

Now I'm learning how not to do them.

That's turned out to be a skill in itself, and I'm not good at it yet.

Because there are two completely different questions hiding in here.

The first is what work can I hand over.

The second is what work do I actually still want to do myself.

Most people, me included, treat those as the same question. They're not. The fastest way to finish something isn't automatically the best way to spend your time on it.

And that's the thing I keep coming back to, because my own role is shifting. I'm not the one laying the bricks anymore, I'm the one describing the building. When that happens to any kind of work, it usually stops looking like the job it used to be.

That's not new either. When steam powered machines first took over manual work, a lot of jobs didn't disappear. They changed shape.

People moved into operating the machines, maintaining them, and designing ecosystems around them.

I think AI might be doing something similar to work now. Instead of writing every line, I'm learning to describe systems, coordinate agents, judge what comes back, and decide what's worth building.

Though I want to be careful here, because that comparison is a double edged sword. Machines doing the heavy lifting brought real benefits, but it also made people redundant and changed how they lived.

Honestly, I don't know how AI plays out, and anyone claiming to is guessing.

What I do think is that we're going to have to shift from the skills we've spent years building into a new set. And the one that matters most might be judging whether the output is actually any good.

I still want to build stupidly ambitious things. That's the whole point. I can now attempt ideas that would've been completely out of reach a few years ago, and I'm not giving that up for anything.

I just don't want the building to turn into babysitting.

WHAT I'M TRYING NEXT

For Week 16, I'm going to see if I can build one AI agent to manage the others.

Something that can assign work, run tests, check what's been built, and only bother me when there's a decision I actually need to make.

Sounds complicated.

Probably is.

But that's the funny thing about all of this.

At the start of the week, my barber was absolutely buzzing because AI had taken one annoying little task off his hands.

Meanwhile, I've been using AI so much that I've somehow invented an entirely new annoying task.

Supervising the thing that's supposed to be doing the work.

So perhaps the strange art of doing nothing isn't about sitting back and letting machines handle everything.

It's figuring out where you're still needed, and where you're just getting in the way.

And judging by the amount of time I spend pressing Enter, I've still got quite a bit to learn.