Let me be clear about something before I say anything else. I love AI. I use it every day. It has genuinely changed how I work in ways I didn't think were possible a few years ago. I've built tools with it, connected it to business systems, used it to prototype faster and think bigger and move with a confidence I didn't always have when I was working alone.
So this is not a doomer take. This is not a designer shaking his fist at the technology threatening his livelihood. This is an honest assessment from someone who uses these tools constantly and has run into the same wall enough times to have an opinion about it.
AI is remarkable at a lot of things. Turning creative vision into usable creative output is not yet one of them.
The gap between the picture in my head and what comes back
Every designer knows the feeling of having something fully formed in their mind before a single pixel is placed. A color temperature. A feeling. A kind of confidence the design should project. The negative space that makes everything breathe. The way the type should sit on the page like it was always supposed to be there.
That's taste. It's specific and it's personal and it's accumulated from years of looking at things and knowing instinctively what works and what doesn't.
When I describe that to another designer, they get it. Not perfectly, not immediately, but they can ask the right questions, push back on the right things, and eventually we arrive at something that resembles what I was seeing. The conversation is part of the process.
When I describe it to an AI image generator, what comes back is usually a confident, high-resolution interpretation of the words I used, not the idea behind them. And those are very different things.
I've tested this more times than I can count. I'll spend real time on a prompt. I'll describe the logo I'm imagining, the weight of the mark, the relationship between the type and the symbol, the industries it should feel adjacent to without looking like it belongs to any of them. I'll get specific about what I don't want. I'll reference visual languages without naming them directly. I'll try every variation of the prompt I can think of.
And what comes back is almost always something I would describe as a confident miss.
Not bad, necessarily. Sometimes technically impressive. But fundamentally disconnected from the thing I described, in ways that are hard to explain but immediately obvious to anyone with a trained eye.
[Example: prompt vs. output comparison here]
It's not a prompting problem
This is the part that matters most, and it's the part I want to push back on hardest.
The common response when AI creative output falls short is that the prompt wasn't good enough. Learn to prompt better. Be more specific. Use the right language. It's a skill like any other.
I've heard this enough times that I went out of my way to test it seriously. I wrote detailed, structured, technically precise prompts. I used reference terminology that any trained designer would immediately understand. I iterated the same concept through dozens of prompt variations, adjusting one variable at a time like I was running an actual experiment.
The results got marginally better in some cases. They did not close the gap.
The issue isn't that AI doesn't understand the words. It clearly does. The issue is that creative vision isn't made of words in the first place. It's made of intuition, judgment, reference points, aesthetic history, and an accumulated sense of what a thing should feel like in the world. You can approximate those things in language. You cannot fully transmit them.
What AI does extremely well is pattern recognition at massive scale. It has seen more design than any human ever could, and it can identify and reproduce what's common, what's popular, what resembles things that already exist. That's genuinely useful for a lot of tasks.
But originality, the kind that comes from a specific person's specific taste making a specific decision that runs counter to what everyone else would do, that's a different thing. And I don't think it's a prompting problem. I think it's a fundamental limitation of how these systems currently work.
[Example: same concept, multiple prompt iterations, output comparison]
What AI actually does well creatively
I don't want to overstate the limitation, because there's a real version of this that makes AI sound useless for creative work and that's not true.
Where AI genuinely helps me is in the early, messy, directional phase of a creative problem. When I need to explore a range of possibilities quickly, when I want to see ten directions before I commit to one, when I need to check whether an instinct I have is worth pursuing, AI is genuinely useful for that. It's a fast, tireless brainstorming partner that doesn't have ego about its ideas.
It's also good at execution tasks where the creative decisions have already been made. Resizing, reformatting, filling in, extending. The stuff that used to eat hours of a designer's day and required real skill to do well but not necessarily real creative judgment.
Where it struggles is in the middle, the translation layer between a strong creative vision and a finished artifact that actually embodies it. That middle part is, arguably, most of what design is. And right now, that's still on the human.
Why this matters for the future of design
There's a version of the "AI will replace designers" argument that assumes creative output is a production problem. That if you can generate enough variations fast enough, one of them will be right.
My experience tells me that's not how good design works. Good design isn't about generating options until something sticks. It's about having a point of view and making decisions from that point of view, again and again, until everything in the work is intentional. That process requires judgment that AI can assist but not yet replace.
What I think will actually happen, and what I'm already seeing happen in my own work, is that AI raises the floor significantly. The gap between a designer with great taste and no time and a designer with average taste and all the time in the world is getting smaller. Execution is becoming less of a bottleneck. That's good for the field overall.
But it raises the ceiling too, in a way that I don't think gets talked about enough. Because when execution is less of a constraint, taste becomes more of a differentiator. The designers who know what good looks like, who have the judgment to evaluate what AI produces and push it in the right direction, are going to be more valuable, not less.
You can't outsource the eye. Not yet.
[Example: AI-assisted workflow vs. AI-replaced workflow, side by side]
The part that actually makes me feel good about this
I'll end on something that surprised me when I sat down to write this.
There's a version of the AI creativity conversation that feels like a threat and a version that feels like a clarification. For me it's mostly been the second one.
Every time I've described a creative vision to an AI tool and gotten back something that completely misses the point, it's reminded me of something I think designers sometimes forget when the anxiety about automation gets loud. The thing you have, the specific accumulation of taste and judgment and visual instinct that makes you good at this, that is not replicable. Not yet. Maybe not ever in the way that actually matters.
AI has seen more design than I ever will. It can generate more options in an hour than I could in a year. And it still can't see what I see when I close my eyes and imagine what something should be.
That's not nothing. That's actually everything.
What did you think?
Share with friends