AI didn’t do what we expected. It did something better.

Not what we expected. Not what you're probably hearing from everyone else.

We assumed AI would speed things up and help with the tedious stuff, the scheduling, writing meta descriptions, the repetitive tasks nobody wants to do. And it does. But what no one wants to talk about is how AI elevates the work itself. To be clear: it cannot create bold, meaningful work on its own. But by helping us build a richer foundation—what we create on top crosses into territory we couldn't reach before.

There's a lot of noise in the industry about how close AI can get to 100% accuracy—and for things like autonomous vehicles or medical diagnostics, that's the right metric. But in the design world, what matters more is whether you can create something genuinely distinct: expressing the brand and solving problems in a way nobody else has. That's where AI hits a wall. Trained on what already exists, it lacks the judgment, taste, and nuanced understanding of human emotion that genuine distinctiveness requires.

So the better metric for design is: how strong a foundation does it build? And what we've found, again and again, is that while the accuracy and relevance of AI output sits somewhere between 50–80%, it's exceptional at establishing a base to work from, surfacing strategic possibilities, generating draft wireframes, structural code, and directional copy that shapes a page's story before design locks it in. By starting with that foundation and repeatedly redirecting and refining its outputs, we consistently reach work that punches above what we could previously deliver. AI raises the floor. And a higher floor is how you raise the ceiling.

What to watch out for: AI can make weak thinking look convincing. Unlike a bad design, which announces its incompetence, AI-generated work can look polished and considered while the underlying logic still has real holes in it. Stay just as critical of it as you would anything else, maybe more so, precisely because it looks polished.

Most people use AI to generate ideas. We use it to punch holes in them.

When we're developing a concept, we give AI our thinking, the brief, the direction, the reasoning, and then ask it to find the weaknesses. To adopt the perspective of a skeptical stakeholder or a user who doesn't care about whether it’s clever. It's uncomfortable in the right way. It surfaces the places where we've talked ourselves into something rather than actually figured it out.

One technique worth knowing: in Claude Code, you can run multiple AI models simultaneously and have them critique each other's ideas. You get genuine disagreement, not one model's perspective but several, in tension, and then a synthesis of where they converge and where they don't. It's become one of our most reliable ways to stress-test a concept before we've committed to it.

What to watch out for: Even when you’re trying to argue with AI, it’s still wired to please. It reflects your assumptions back at you with more confidence than they often deserve. A practical fix: run the same prompt in an incognito window, where it has no memory of your conversation. You'll often get a noticeably different answer, and the gap between the two tells you something.

There used to be a significant gap between conceiving something and seeing it function. A concept would live in a deck or a static mockup for weeks before anyone could interact with it, and in that gap, a lot of bad assumptions often went unexamined. You'd talk about how you think a user would move through it without being able to see it and test it in action.

That gap is now very small. In many cases, we're shifting our design and testing process to go straight from idea to working prototype—not polished, not precious, but functional enough to put in front of someone and learn something real. We're skipping the part where we talk about what we think will happen and going straight to finding out. We use it, we watch others use it, and within hours we know what's actually true about it, and we can start refining from there, based on what's real rather than what we assumed.

What to watch out for: When prompting AI to build a prototype, stay focused on what you're trying to learn, not building everything and the kitchen sink. It's easy to get pulled into rabbit holes with AI, chasing a more polished UI or adding cool features…and surface hours and hours later, defeating the purpose of moving fast.

When building becomes more fluid, the process reorganizes around it. The old model—a phased approach of strategy, design, development, then review—was built on the need to define and separate each step before moving forward. It was linear by necessity.

AI allows ideas to take shape much earlier, turning strategy into something that can be explored through real outputs, not just defined upfront. Instead of moving step by step, work becomes more iterative—strategy, design, copy, and build evolving together in continuous loops rather than clean handoffs.

That changes how teams work and what clients experience. It also changes what gets made. When ideas can be explored and refined more easily, it becomes possible to go deeper, serve more specific needs, and pursue opportunities that previously felt out of reach.

Learning to actually partner with AI, not just prompt it and accept whatever comes back is harder than it sounds. It required us to unlearn some habits that were deeply ingrained and to hold two things at once: a genuine beginner's openness, and enough experience to know how to turn the output into something that leapfrogs the industry.

We don’t always get that balance right. But we’re doing the best work in our studio’s history—and we’re just getting started.