By Andy Siemer, CEO
How do you get a designer who's spent years honing her skill to even try the thing she's already decided is beneath her? I hear some version of this from most creative people we work with, and it's a fair pushback. Designers, marketing people, anyone who's spent years getting good at a craft — they hear "AI" and what they actually hear is "the thing that's about to tell you your years didn't matter." So they push back. I've worked for years to hone my skill, and I'm better than a machine. And they're right. They are better than the machine, at the parts that actually require them.
That's the hill. Arguing about it doesn't work. So I don't argue about it — I ask them to try it on one real request.
I told them straight: "it doesn't have to be thought of that way. Think about it as a different kind of paintbrush or a different kind of tool." You still decide what's good enough. You still hold the brush.
Marcus, a client of ours running a custom apparel and promo shop, has a designer on his team — we'll call her Jess in this telling — who'd never touched an AI art tool. Her boss gave her a test brief pulled straight from a real client: Miami skyline, pastel colors, nighttime, the city name arched above it. I assume most shops get some version of that request every week — correct me if yours doesn't. She typed it in. Four usable options came back in seconds.
The only real snag was that the text across the top wasn't the client's brand. So she clicked on it, removed it, and swapped in their actual logo. That part took her about a minute. She didn't need to learn prompting or a second tool. She typed the brief in plain words and got four things back she could actually show somebody.
I want to be honest about where this stops, because I think that's the part people skip when they're trying to sell you on a tool. The output wasn't perfect. It never is on the first pass. And there are jobs an AI tool isn't going to do for you at all — a client sends over a rendering of their building, and somebody on Marcus's team still has to turn that into a full custom t-shirt design by hand. That takes hours, and it stays that way. Nothing about a fast skyline mockup replaces the person who can take a building rendering and make it wearable. The tool is good at the fuzzy front end — get me something in the neighborhood of right — and useless at the part where the client's actual building has to survive being printed on a shirt.
The honest pitch is the one I gave on that same call: "worst case scenario, use it for inspiration, gathering, mood board creation, get yourself there 80% of the way, and then you can tweak it. So you go from a six-hour project to a one-hour project." That's the number that made the case, and it's not a bigger claim than that. It's mood-boarding faster. The first draft shows up in seconds instead of by the end of the day.
Here's the part I'd push back on if you only heard the first half of this. "We can do anything" is not the same sentence as "run everything through AI." Those get treated like they're the same idea, and they're not. One of our team and I were talking through this recently. The way we put it internally: "everything can start as AI, but we want to get off of AI where possible because running workflow through AI is more expensive than running workflow where it's just scripted and just works." When something proves out and the behavior is stable, we move it off the model and into a plain script. A script that already works is cheaper to run every single time than a model call that has to think about it again. There's no reason to keep paying a model to re-solve a problem you already solved.
So the sequence, in practice: start with AI wherever the problem is fuzzy — where you don't know the shape of the answer yet, mood boards and first drafts, the stuff that used to eat a whole afternoon. Let a human look at the output and decide what's good enough. And then watch for the moment that fuzzy thing stops being fuzzy. If Jess starts using the same handful of prompts for the same handful of requests, that's a pattern, and a pattern is exactly the kind of thing you can write once and stop paying a model to redo. Once a piece of that work turns out to be the same shape every time, graduate it off the model. Write the script. Save the AI budget for the next fuzzy problem, because there's always another one.
None of this requires Jess to believe AI is as good as she is. It requires her to try it on one real brief and see what comes back in seconds instead of hours. That's a much smaller ask than "trust the machine." It's just: pick up the different brush and see what it's actually good at. She still decides what's good enough.
First pass is a list of the fuzzy-front-end requests your team gets most — you cross off the ones that are real craft. Start the conversation.