Creative process is exactly the step that Generative AI seems to skip. Pop in a prompt, get an email or a flyer out. No muss, no fuss, no actual process at all.
This is, of course, a recipe for slop: a blog post that doesn’t say anything, a visual that doesn’t do any work beyond taking up space on the page. This is the stuff that is quite rightly turning many people, especially those who value human creativity, against Generative AI.
And as for the economics: if the stuff you get out of Generative AI doesn’t move the needle — excite, persuade, intrigue, enhance — then you’ve gotten zero value from that webpage, flyer, or campaign, however quickly or easily you made it. If it’s not converting and selling, it’s not doing its job.
One of the very weird things about being a creative worker in this moment of burgeoning AI use is that we would never, ever want to produce something that “looks like AI made it.” But at the same time, as innovative, creative thinkers and makers, we’re extremely interested in figuring out how we can use Generative AI. What does it make possible? What can we do with this powerful new tool?
So when we use Generative AI, it’s often as part of the process (or handing over the boring, repeatable chores) — not to conjure up the actual final thing we’re making. It adds value or acts as a force multiplier in the exploration it facilitates, the perspective it brings into focus, the blockers it erases, the dead ends we can see coming more quickly.
As we play with Generative AI, we grow our understanding of where it’s useful and where it’s really not. We would never let it interfere with the quality of work that we ship: if anything, it lets us be even more rigorously precise in executing a particular brand.
But it’s our creative minds — our taste, experience, and judgment — that are the movers. The AI is one more tool in the toolbox.
{{divider}}
No more blank page: Gen AI gives you something to react to
Lots of us use Gen AI to simply put something on the page.
The value of this cannot be overstated. Every writer or designer who has ever lived has suffered through starting a project and realizing their mind is blank. We don’t expect to get anything good on a first try with Gen AI. But once there’s something there, our judgment can get to work. There’s a very strong itch to fix something terrible, and that right there is a starting point.
Design Lead Kaylee Green is an intuitive, exploratory designer. She really hates a blank page. She needs something to react to, or a conversation to set ideas spinning. Sometimes she turns to other designers (or even writers) to get the creative juices flowing. Sometimes she turns to Gen AI.
For her, Gen AI becomes an extension of thinking: a way to surface an idea, look at it, reject it, reshape it, and figure out what she actually believes. For one client, she tried out AI-generated wireframes that ended up being too blocky and sterile for the brand. None of them became the final design. But they still helped. They made clear what the site should not be, which gave her a sharper sense of what it needed to become.
Faster experimentation and iteration lets you get past wrong ideas faster.
Design Lead Stetson Finch uses his array of favorite tools to move from broad ideation toward more specific execution. While developing the Huntress art direction, he had Claude build rough forms in Blender. He didn’t use what it made. The value was seeing the idea rendered enough to figure out what he wanted to change next. Seeing the idea take shape opened his brain to the next step of the art direction. He had to get part of the way there to see the next leap forward.
As a Brand & Interactive Writer, I use Gen AI similarly in writing. I know the brief, the strategy, and the constraints long before I know the sentence. So I use it to generate options, rough drafts, alternate structures, or bad ideas quickly. Then I can start doing the part I’m really good at: comparing, diagnosing, cutting, recombining, and deciding what actually works. Or sometimes stepping back to reframe the question entirely.
{{divider}}
Show me where I’m wrong: Gen AI manufactures productive friction
Sometimes the most useful thing Gen AI can give us is resistance.
Brand & Interactive Writer Ware Carlton-Ford recently developed a new verbal workshop. He had a strong sense of what he wanted it to accomplish and the experience he wanted to create. He drafted an outline and had ideas for crafting the flow of the learning. But before he built it all out, he paused. It was time to poke holes in his work. To consider what could go wrong.
What had he not thought through? Where was he relying on what he expected people to do, and missing possible responses? How could he structure the workshop to get people writing, when writing on demand can be extremely intimidating (especially in front of your boss)?
Ware asked Claude to help him answer these questions — and to come up with more potential failure modes. He kept pushing until Claude ran out of useful questions.
He compares the process to using video while learning a physical skill.
Ware used to be an acrobat (yes, really). He knows, from painful experience, that you can be absolutely convinced that you’re moving your body the way the coach told you to — until you watch the tape and realize you’re not. The outside view reveals something your internal experience can’t.
This is one of Ware’s preferred ways to use Gen AI, as “a camera outside my own perceptions.” Not because the machine understands the problem better than he does, but because it can give him another angle from which to examine his own thinking.
He calls it a “shake things loose box.” When he finds himself circling the same ideas, he can use Gen AI to introduce something unexpected: a question, a structure, a possibility that wasn’t otherwise going to pop out of his head that day. Then he decides whether any of it is actually useful.
Claude gives Ware something new to push against — and sometimes that friction is exactly what gets his own thinking moving again.
Brand Strategist Jordan Keller takes pushback a step further, building it into his process as an essential step.
He’ll load in frameworks he trusts — positioning theory, category-creation criteria, definitions he’s refined over time — and then give Claude his ideas and drafts for this project. He tells Claude straight up to be ruthless. Tell him he’s wrong. Come in trying to prove what doesn’t hold up. He wants the fight.
When he’s using Gen AI to explore ideas, Keller adds friction into the process deliberately. Left to its own devices, Claude, like all of these tools, is often far too agreeable. That’s how it’s built. It will nod along with your premise and help you elaborate it, even when what you really need is someone to poke viciously at any weak spots. Keller brings his own well-honed strategic point of view. The tool helps him figure out whether the logic beneath it can take a hit. Claude is his sparring partner.
Keller isn’t asking Claude what the strategy should be. He’s using it to make his own strategy harder to fool himself about.
Keller also knows when not to ask Claude anything yet. Early in a project, he imposes “speed limits” on himself: read the client materials, absorb the problem, sit with the information before inviting AI into the synthesis. He has to own the information, build a framework in his mind that captures the essential nature of this client. Exploring with Gen AI can be blazingly fast: too fast. If he gets ahead of his own understanding, his ability to judge the outputs gets shaky. He hasn’t “earned the making of meaning.” This intentional slowdown ensures that he can defend every decision and recommendation, because while the tool may have helped him get there, the guiding mind is his.
Pushback by itself isn’t enough, of course. It’s useful because both Ware and Keller have the expertise, taste, and judgment to evaluate the feedback Gen AI can show them. They are the ones who know when the final product is actually good: ready for client eyes.
{{divider}}
Keep the context in the room: Gen AI as memory on tap
Creative work doesn’t happen in a vacuum. By the time we’re making a recommendation, writing a headline, or choosing a visual direction, there may be weeks of strategy, interviews, research, feedback, and tiny decisions behind it.
Gen AI is massively useful at keeping more of that material available while we work.
Stetson feeds his tools kickoff documents, meeting transcripts, strategy, client feedback — essentially the accumulated context of the project. Of course, we use the appropriate privacy and data controls when working with client materials. This growing slew of separate items transforms into a treasure trove of rich information that he can query and manipulate. He can ask for the exact thing a client said three meetings ago, surface the strategy point relevant to the decision in front of him, or check whether the rationale he’s building actually lines up with what the team agreed to earlier.
I work similarly with language. Research, transcripts, competitive analysis, product information, positioning, feedback — all of it can stay close at hand while I’m working on the next page or paragraph. That means I’m not limited to whatever happens to be sitting at the front of my brain that morning.
And I can test new work against a surprisingly complex matrix of what we’ve already established. How does this message line up with the client’s archetype, attributes, positioning, audience, and defined brand voice? Where does it reinforce the system? Where does it create tension?
Then I decide what that tension means. Maybe the copy has wandered out of bounds. Or maybe it reveals a facet of the brand personality that deserves to stay — something that adds complexity and nuance rather than inconsistency.
Gen AI is useful here because it can keep all those different elements in play at once and surface relationships between them. It doesn’t resolve those relationships for me. That’s still a creative judgment.
{{divider}}
Make tacit expertise explicit: Gen AI to translate artistic choices into encoded rules
A lot of creative expertise grows organically, over time and through experimentation. You know something is wrong before you can necessarily explain why. The curve feels too tight. The sentence is dead on the page. The composition is technically fine but utterly boring.
Gen AI is incredibly useful at forcing that instinct into articulation.
Kaylee ran into this while building a visual system inspired by trains, maps, and movement. She knew what the lines should feel like: how they should bend, overlap, split, and move together. She wanted to build a tool for the client to use to generate elements of this visual language, to use across their website. But in order to do this, she had to translate her design instinct into language that Gen AI could understand.
The first tries were very bad. (This is often true of creative work of any kind.) So she iterated. She told Claude: This is too chaotic. Those curves are too tight. These lines shouldn’t cross that way. When her plain-language feedback wasn’t getting the result she wanted, she asked ChatGPT to help translate what she meant into language Claude’s code could act on, playing the models against one another.
Bit by bit, the tool taught her how to teach the tool.
I use a similar process in writing for clients, but the loop happens inside the drafting itself. I use Gen AI to generate language, and then I critique it, edit it, reject it, redirect it, and explain why something works or doesn’t. Over hundreds of tiny decisions, I’m effectively teaching the tool my editorial judgment about this particular client. I’m figuring out how this brand personality expresses itself through language while, at the same time, teaching the AI about my editorial decision-making process.
Kaylee says that working with Gen AI often feels like “talking to myself.” I have that experience as well: It reflects back to me what I put in.
Then every so often, I stop and ask it to look back across the work we’ve done together. What patterns have emerged in the choices I keep making? What have I repeatedly changed? What rules seem to explain why one version worked and another didn’t?
That meta-analysis surfaces judgment I’ve been applying instinctively. I then examine the rules, correct them, sharpen them, and add the useful ones back into the instructions. The next round of drafts starts from a better place.
Draft, judge, revise, extract the rule, teach it back, draft again. Lather, rinse, repeat.
Eventually, what began as hundreds of small editorial choices becomes a much more explicit model of how this client writes. And that has a practical payoff: I can turn the accumulated judgment into a tool the client can use to produce surprisingly strong first drafts on their own.
{{divider}}
The workflow is part of the experiment: The only constant with Gen AI is change
Sometimes the problem isn’t the prompt. It’s the tool.
While building an image-to-drawing generator for a client, Generative AI Experience Designer Thato Maselela first tried to make Claude handle the transformation. The results weren’t strong enough. So instead of endlessly rewording the prompt, he tested other models, found that Gemini handled the image work better, figured out how to connect Gemini through an API, and used Claude to build the interface around it.
The final workflow used different tools for different jobs.
That kind of experimentation is part of the work now. The models change quickly. Their strengths overlap, diverge, and leapfrog one another. Getting good at Gen AI means learning not just how to ask better questions, but how to recognize when you’re working in the wrong tool — and when the workflow itself needs to change.
Keller applies the same mindset to his own process. At the end of a project, he’ll sometimes ask Claude to look back across everything they did together: Given where we ended up, how should an expert have briefed you at the beginning? Where did we waste time? What could I do differently next time?
It’s a postmortem on the AI-assisted process itself.
The goal isn’t to arrive at one perfect workflow and lock it down forever. It’s to keep refining how we use the tools as both the tools and our own understanding evolve.
{{divider}}
Gen AI is another tool to think and create with
Gen AI is most useful when it gives creative people more to think with.
Something to react to. Something to argue with. A way to bring more context into the room. A mirror for our own instincts. A method for turning repeated judgment into clearer rules. A constantly changing set of tools that we keep learning how to use better.
Gen AI reflects back what we put into it: our references, our standards, our questions, our criticism, our sense of what good looks like. The better and more specific our own thinking becomes, the more useful the tool becomes in return.
None of that replaces taste, judgment, expertise, point of view, or the human work of making meaning. Gen AI can generate, retrieve, challenge, remix, and encode. But someone still has to decide what matters, what belongs, what is true to the brand, and what is good enough to ship.
What it can do is extend those abilities. It gives us more surfaces to think against, more ways to test an idea or to see what we know and where we’re still guessing. And sometimes, when all of that clicks, the results are genuinely surprising: faster exploration, sharper thinking, stronger systems, better work.
That matters because the creative process — while a pleasure in itself — isn’t an end in itself. Not when we’re at work. All of this experimentation is in service of what we ultimately hand to our clients: clearer strategies, more distinctive brands, more usable systems, and work that holds up when it leaves our hands.
