200 designers, drawing each other. First with our good hand, then with the other one. My good-hand portrait was fine but forgettable. My left-hand attempt was a lopsided disaster, and of course it's the one that got a laugh, and, weirdly, the one I liked more.
That was Brian Collins closing out three days of The Design Conference in Brisbane, and it pretty much summed up everything I'd just sat through: the polished version isn't always the one that matters.
I'd flown over with a question our design team keeps chewing on. As AI takes on more of the actual making, where does the designer's value concentrate? Our hunch has been that it lives in craft, taste and judgment; that we move from doing the work to directing it. I went looking for one decent reason to believe that and I came home with a carry-on bag full of them.
The stuff that stuck
The conference opened with typographic illustrator Gemma O'Brien, who uses AI like any other tool but keeps coming back to a big, chaotic notebook of ideas, including the failures she loves digging up and remixing. In her hands, craft isn't the polish at the end. It's where the whole thing starts.
Then Marcel Ziul handed me the line I've quoted a couple times since. "Let chaos speak, but don't let it run the room." Which is basically the job description for working with generative tools right now.
But the talk that I can’t shake came from Peter Barber, a designer building AI cybersecurity tools. His point was simple. As AI makes everything faster and more frictionless, our job isn't to sand off every last bit of friction. It's to put it back exactly where a human needs to stop, understand and decide.
A machine can be fast and it can even be good. What it can't be is accountable. "AI cannot hold the weight of someone's future and know what that means," he said. He called that space the human margin, and unlike the skill of asset generation, it doesn't shrink as the models get better.
Type designer Eleni Beveratou sharpened it further. AI is trained on what already exists, so it drifts towards the average, the safe, the already-done. Reaching past that default is the craft. And Shruthi Manjula Balakrishna gave me the most useful reframe of the week: hand the boring, mechanical mahi to the machine so that you've got room to sit with the hard questions.
The one in the inflatable costume
The best surprise was how much of the conference argued for joy. Alfredo Enciso's talk, "The Economy of Fun: Why Personal Projects Matter," was a parade of the silly, delightful and beautiful things his studio makes alongside the paying work. He delivered part of it in a full inflatable costume, to make an entirely serious point: fun isn't a break from good work, it's the engine of it. You don't, as he put it, have to be serious to do serious shit things.
By the end, the same idea had reached me from a type foundry, a motion studio, an illustrator and an AI specialist. The part that stays ours is judgment - deciding what's worth making, knowing when to refuse the average, and being willing to put your name to the result. The tools are astonishing, and getting better every week. Not one of them can do that bit.
Which raises a more uncomfortable question. If the human margin is real, are we actually working in it? Or are we designers quietly letting the tools set the defaults while we tidy up behind them?
What I'm taking back to my team
Less a summary, more a dare:
- Point our agents at the grunt work, then actually spend the time we win back on judgment, not just more output.
- Find our accountability moments, the calls a human has to own, and design them in on purpose.
- Build a habit or two that shoves our work off the first default the tools reach for.
- Guard time for the odd, unbriefed, slightly-for-fun projects. The best studios on that stage treat them as fuel, not a treat.
- Leverage your unique perspective. Our lived experience and the way we work with other humans is part of the craft, too.
Craft was never the thing at risk here. The three days I spent in Brisbane left me pretty sure it's what matters most.
So, the real question for us isn’t "what can AI do now?" But when a machine can make almost anything, what do we choose to make?