(This article is an extended version of my talk at GrafConf 2025)

AI is the kind of thing everyone wants to describe in superlatives or in apocalyptic terms. Either revolution, or ruin. Me? I just feel like saying: I don’t know.

And maybe that “I don’t know” is the most honest starting point. Because if AI is interesting at all, it’s in the way it disrupts—how it scrambles our established points of reference. Sometimes it breaks things, and that breaking can turn out to be more fertile than perfect execution.

The AI world is obsessed with efficiency. The mantra is always: “shorten processes,” “cut costs,” “speed up results.” And sure, I get it—that’s the logic of business. But creativity—the entire field where the point is to generate something that touches people—doesn’t emerge from acceleration. It comes from friction. From the fact that things don’t fit, don’t work, take time, and involve uncertainty.

Meanwhile, the AI sold by LinkedIn messiahs is more like an organized tourist trail: guide included, viewpoints preselected, a café halfway through. Everything smooth, safe, optimized for “user experience.” The problem is that the most interesting views are always off the trail. And you usually get there with ticks on your legs and mud on your boots, not with a certificate in “creative process efficiency.”

Every project, at its core, is a network of decisions—not just “what” and “how,” but also “whether at all.” And the real value often emerges in those moments of hesitation: inefficient, costly, risky moments. The places where it’s easy to screw up.

AI, in its automation mode, cuts those places out like a surgeon. What’s left is a product: smooth, repeatable, sterilized. Like plastic flowers: they don’t wither, they don’t make a mess, they look nice and predictable… and they’re dead.

I prefer a different approach. Instead of asking AI for the “best solution,” I’d rather ask it something that derails it. Provoke it into tripping. That’s when the strange dance begins—it stumbles, I stumble along with it, and in that stumbling something new opens up. This isn’t about fetishizing error—it’s about practicing disruption. Because in a world obsessively smoothed and polished for the sake of “user journeys,” the error becomes an act of resistance. It’s the moment a crack appears in the system—and I want to step right into that crack.

AI as a partner? Yes—but not as a content factory. More like an opponent in a tug-of-war. I pull one way, it pulls another—and in that tension, something interesting starts to happen. Inspiration doesn’t come when the system runs as expected; it comes when something slips, when there’s a snap, a dissonance.

It’s like a conversation—not the polite one where everyone agrees, but the kind where someone suddenly throws out an absurd line, and you realize you have to build a whole new train of thought. Interaction with AI isn’t clicking “generate” and passively waiting. It’s a game. The question itself can be an artistic gesture. And the answer—a raw material you break apart, stretch, distort.

The problem is that AI often narrows contexts rather than expanding them. The internet is drowning in photorealistic images that no one needs. We’ve thrown ourselves into a global race to churn out endless, interchangeable content. AI slop: the surplus of things that look fine, correct, aesthetically pleasing… and could just as well not exist at all.

That’s why we need a different kind of thinking—skeptical, full of disruption. Not the kind that follows yet another LinkedIn framework, but the kind that drags us into places we’d never think to go ourselves. It’s not about AI doing our work for us—it’s about AI pulling us into strange, uncharted territory.

I’m not against AI. I’m against the comfort that AI so eagerly sells us. Because comfort in creativity is poison. Creativity isn’t supposed to be comfortable—it’s supposed to be awkward, rough, uneven. The most interesting things happen when the system—human or machine—doesn’t work according to plan. Glitches, absurdities, the stuff that doesn’t fit—these aren’t waste products. They’re the fumes that, handled right, become fuel.

Maybe today—at the height of generative convenience—what we need more than polished results are consciously worked-through missteps. The kind that carry energy, that provoke further searching. Because the only guarantee that something will really be yours is that along the way, you’ll ruin it a little. All the things that had been so neatly optimized and carefully prepared.

So don’t ask: how can AI help me optimize my process. Ask: how can it knock me out of it.