Use AI to Build Ideas That Don't Need It
I’ve been watching something quietly unfold over the past year, and it’s making me uneasy. Not because AI is taking over creative work — I’ve already said my bit on that. But because I’m watching smart, talented people build their entire creative process on top of a thing they don’t own, can’t control, and have absolutely no guarantee will still exist in its current form in eighteen months.
Token costs are rising. Enterprise agreements are being renegotiated. The tools that felt free and frictionless in 2023 are developing price tags that make you wince. And the access creatives assumed was a permanent feature of the landscape is starting to look more like a promotional period.
Here’s the uncomfortable question nobody’s asking out loud: if you removed the AI tomorrow, would your ideas still stand up?
The dependency trap nobody’s talking about
When I was building Big Drop, we were working within some of the tightest constraints I’ve ever encountered in a creative brief. Zero alcohol, mass-market ambition, a category that didn’t really exist yet in any meaningful way — and a very specific story to tell about why this beer belonged in your hand rather than just not having a beer.
The constraint was the brief. The constraint was the creative.
We couldn’t rely on the halo of booze culture, the nostalgia of legacy brands, or the easy shorthand of “this is what beer looks and sounds like.” We had to think from first principles. Every single decision — every word, every visual choice, every moment of brand behaviour — had to earn its place. Nothing could coast.
That experience taught me something I’ve returned to over and over: the best ideas are the ones that survive the removal of every crutch.
AI, used badly, is a crutch. A very clever, very articulate, very convincing crutch — but a crutch all the same.
The bit people get wrong
There’s a version of using AI in creative work that I think is genuinely brilliant, and it’s not the one most people are doing.
Most people are using it as a production tool. Feed it a brief, get a first draft, iterate until it’s good enough, ship it. Faster. Cheaper. Covers the gaps.
I get it. I understand the commercial reality. But what you’re building in that scenario isn’t your thinking with AI assistance — it’s AI thinking with your name on it. And when the access changes, or the model shifts, or the quality dips, you’ve got nothing in the bank. No creative muscle memory. No accumulated instinct. No ideas that are irreducibly yours.
The version I find actually interesting is using AI to go further with your own thinking — not instead of it.
Start with the hard bit. The bit that actually hurts. Sit with the problem, develop a point of view, sketch an argument. Get messy. Get confused. Work it out with whatever rough tools you have in your head, on paper, on a whiteboard, in a voice note while you’re walking the dog.
Then bring in the AI.
Use it to challenge what you’ve built. Ask it to find the holes. Make it argue the opposite case. Use it to stress-test the assumption buried three layers down that you haven’t even spotted yet. Let it generate ten versions of the same idea so you can look at them all and know, immediately, which one is yours.
That’s AI as a thinking partner rather than a thinking replacement.
The output still needs to stand entirely on its own two feet when you’re done — and if you’ve done the first part properly, it will.
Why this matters more now
I said token costs are rising. I’ll say it again, because I think people are genuinely not paying attention to what’s coming.
The creative AI tools that felt like a permanent democratisation of capability were, in many cases, venture-subsidised experiments. The businesses behind them needed users before they needed revenue. That phase is ending.
What replaces it won’t destroy AI — these tools are too useful for that. But it will stratify access. Enterprise clients and large agencies will keep paying. Independent creatives, small studios, and early-career practitioners will face a harder calculation. The workflow you built assuming free or near-free API access may cost you real money to maintain. And the creative muscle you didn’t bother building because the tool was doing it? That’s not going to grow back overnight.
I’m not saying don’t use it. I’m saying: build the thinking first. Own the idea before you use the tool to polish it. Develop a creative practice that can survive without it — and then use it to go further than you’d go alone.
The best use of any powerful tool is to make yourself more capable, not more dependent.
The question worth sitting with
There’s a version of this that’s uncomfortable for me to write, because I use AI. I use it regularly. I use it to research, to challenge my arguments, to work through briefs when my brain is bouncing off walls and I need something to think at. It’s been useful in ways I didn’t anticipate.
But I know — and I mean know, not just believe — that if you took every AI tool away from me tomorrow, my ideas wouldn’t change. They’d be slower to stress-test. Rougher around the edges. Maybe I’d miss a counterargument or two. But the thinking underneath? That’s mine. It was mine before the tools existed, and it will still be mine when the current crop of tools has been replaced by whatever comes next.
That’s what I want for the people I work with and the people I mentor. Not dependency dressed up as efficiency. Not polished output that evaporates the moment the subscription lapses.
Ideas that don’t need the tool to survive. Built with it, maybe. But not dependent on it.
That’s the only kind worth building.
If this resonates, you might also find something useful in The Curator of Good — my earlier thinking on why human taste still matters more than pattern-matching engines, and The Algorithm in the Room on how AI integration becomes distraction rather than efficiency.
