I love AI. Which is exactly why its dark side worries me.

Let me get this out of the way first: I love AI.

It has genuinely changed how I work. The dull bits of building websites, the boilerplate, the “I know how to do this, I just can’t be bothered typing it all out” bits, are largely gone. I ship more, faster, and I get to spend the good hours on the actual thinking. I’m not one of these people quietly hoping it all blows over. It won’t, and I don’t want it to.

Which is exactly why I feel like I’ve earned the right to say the next part.

The sales pitch around AI is so relentlessly, blindingly shiny that nobody’s really putting the downsides on the slide deck. And there are downsides. Not “the robots will kill us all” downsides, that’s mostly sci-fi with a marketing budget. Quieter ones. The kind you don’t notice until you look up one day and realise something’s gone.

So here’s my slightly uneasy list. Not to scare you off it, I’m the last person who’d do that, but to use the thing with your eyes open.

We’re quietly outsourcing human connection

The clever demo is always the same: AI writes your emails, handles your customer chats, drafts your messages, books your appointments. Brilliant. Efficient.

But stack enough of that up and you end up with a world where a machine writes to a machine that summarises it for a human who never really turns up. The thank-you note nobody wrote. The “just checking in” that was actually a scheduled automation. The support chat that’s a bot pretending to be a person named Sarah.

The whole point of a lot of this stuff, the notes, the calls, the awkward little human touches, was that a person bothered. Take the bother out and you’ve automated away the exact thing that made it mean something.

When nobody actually turns up

Follow that thread a little further and it gets properly strange. Today AI drafts the meeting invite. It’s not a big leap to a near future where it attends the meeting for you. You send your agent, I send mine, the two of them have the conversation, and then each dutifully summarises it back to the human who couldn’t spare the half hour. Two people who never actually met, each handed a tidy transcript of a meeting that, in every sense that matters, didn’t happen.

Wildly efficient. Also faintly absurd. At some point you have to stop and ask what we thought meetings, or working relationships, or for that matter people, were actually for. I run a small studio precisely because I like sitting across a table from the people I work with. I’d quite like that to still be worth something in ten years.

We’re outsourcing our own thinking, too

This one worries me more than I expected it to.

There’s a decent chunk of research now suggesting that when you lean hard on AI to do your thinking, your own ability to do that thinking quietly wastes away. Same as a muscle you stopped using. It’s got a name, “cognitive offloading”, and it’s not new, we did it with calculators and sat-navs, but the scale here is different. This isn’t offloading long division. It’s offloading the reasoning.

The skills we’re quietly retiring

And it isn’t only the big, clever thinking. It’s the small, boring competences that quietly hold a mind together, and we’re retiring them one convenience at a time.

Remember handwriting? It’s quietly becoming a party trick.

Take typing. Voice to text is now good enough that plenty of people barely touch a keyboard, and typing is a skill like any other, kept sharp by doing it constantly. Stop, and it goes the way of joined-up handwriting: slower, clumsier, eventually a faff. Nobody will decide to get worse at typing. It’ll just happen, quietly, because we stopped needing to.

Or handwriting itself. When every meeting is recorded, transcribed, translated and summarised before you’ve even left the room, why would you scribble anything down? You wouldn’t, so you don’t. And the odd, underrated magic of writing by hand, the way it forces you to decide what actually matters enough to keep, goes with it.

Or reading. Proper, sit-with-it reading is quietly becoming optional, because why wade through the whole thing when the AI will hand you the highlights? Except the wading was the point. That’s where the understanding lived, not in the summary.

None of these is a tragedy on its own. Together they add up to a mind that’s slowly forgotten how to do the reps.

The danger isn’t wrong, it’s plausibly wrong

I catch myself doing it. Reaching for the model before I’ve actually sat with the problem. And the danger isn’t that the AI gets it wrong, it’s that it gets it plausibly wrong and I’ve lost the sharpness to notice, because I stopped practising. A generation that never struggles with a blank page might also never learn what it’s like to fill one. I don’t know what that does to us. I’m not sure anyone does.

The money and the power are all flowing one way

Here’s the bit that gets waved away as cynicism.

Building serious AI takes an obscene amount of money, computing power and data. Which means, by design, it’s a game only a handful of enormous companies can actually play. The tools are astonishing, but they’re astonishing tools you rent from a landlord who can change the price, the rules, or the whole product whenever they like.

They own the picks and the shovels

And it runs deeper than software you pay for monthly. The companies at the top don’t just own the clever models, they own the picks and shovels underneath the whole thing: the chips, the data centres, the raw compute the entire era runs on. There’s a real, physical scarcity here, and they are sitting squarely on top of it.

The “cloud” is just someone else’s very expensive warehouse.

So it isn’t only that you depend on Big Tech for the shiny tool. You depend on them for the ground the industry is built on. We’ve never been further from being able to shrug and say “fine, I’ll build my own”, because the barrier isn’t skill any more, it’s owning a warehouse of hardware worth more than most countries spend on schools. We’ve spent twenty years handing these companies the keys to search, to social, to the app store, to the cloud. AI is the same story with a much bigger cheque, and this time they’re holding all the hardware too. More power, more money, concentrated in fewer hands than ever, while the rest of us get a monthly subscription and a cheerful “we’ve updated our terms”. Forgive me if I don’t find that entirely comforting.

Privacy was already on life support

Every one of these tools runs on data. Yours, mine, everyone’s. The writing, the questions, the half-formed 2am thoughts you typed into a chatbot because it felt private. It wasn’t.

We’d already made our peace with being the product. AI just raises the stakes, because now the thing hoovering up your data is also clever enough to understand it, connect it, and infer the bits you never actually told it. It doesn’t just know what you typed; it can have a decent guess at what you meant, and what you’ll do next. The line between “helpful and personalised” and “quietly knows more about you than your GP” is thinner than the onboarding flow makes it look. As someone who takes keeping client data safe fairly seriously, that keeps me up at night more than most of this list.

It makes being lazy incredibly easy

AI is a fantastic tool and a fantastic crutch, and the trouble is it’s the exact same thing. The difference is entirely down to you.

Used well, it clears the rubbish so you can do better work. Used lazily, it becomes a machine for producing the appearance of work. Content nobody wrote for nobody to read. Code you pasted in and didn’t understand, right up until it broke in production and you’d no idea why. Answers you didn’t check because checking felt like effort.

The tool doesn’t care which one you pick. It’ll help you cut corners just as eagerly as it’ll help you do something great, and it’ll make the corner-cutting look every bit as polished as the real work. That’s the quiet trap: for the first time, lazy and diligent can produce output that looks identical, right up until the moment it actually matters.

And yes, the jobs question

I’ve saved the big scary one for last, and I’ll be honest with you: I don’t know.

Anyone who tells you confidently that AI either will or won’t cause mass unemployment is guessing, they’re just doing it with more conviction than me. The optimistic take, that every technology shift destroys some jobs and creates others, has history on its side. The uneasy take is that “history’s fine, eventually” is cold comfort if you’re one of the people caught in the eventually, and that this shift is coming for a much broader range of work, much faster, than the ones before it.

What I do know is that “a computer can now do a decent 80% of this job” is a genuinely new sentence for an awful lot of jobs, including plenty that felt untouchable a couple of years ago. Whether that means fewer people or just different work, we’ll find out the hard way, in real time, with real people. Which is not the tidy reassurance I’d like to end on, but it’s the honest one.

So what do we actually do

Not panic, and not pretend. Both are lazy.

I still think AI is one of the best things to happen to how I work, and I’ll keep using it every single day. But loving something isn’t the same as refusing to see its edges. The people who’ll come out of this well aren’t the ones who reject it, or the ones who swallow the hype whole. They’re the ones who stay awake: who keep their own thinking sharp, keep the human bits human, read the whole thing now and again, and treat every shiny promise with a healthy raised eyebrow.

Use the tool. Just don’t let it use you. And every so often, do something the slow, human, inefficient way, by hand, in person, from memory, just to prove you still can.