AI belongs in the process, not the pixels

Every designer I know is now having the same argument, sometimes with their team and sometimes with themselves: how much of this should the machine do? The answers I hear tend to sit at the extremes, full refusal or full surrender. Both are wrong, and both are wrong for the same reason. They treat design as one activity, when it is really two.

There is the process: research synthesis, competitor teardowns, exploring twelve layout directions before lunch, writing the boring variants of microcopy, testing an information architecture by arguing with something that never gets tired. And there is the artifact: the thing that ships with your name on it, the interface a real person uses, the identity a company wears in public.

AI has made the first of these much cheaper. It has made the second much easier to fake. Those are not the same gift.

Where the acceleration is real

In my own work, the process wins are concrete and boring, which is exactly why they hold up. Clustering interview quotes into themes used to cost an afternoon. Now the first pass costs minutes, and I spend the afternoon questioning it. Generating the wrong directions fast, the completeness-first form, the obvious layout, the default everyone reaches for, is genuinely useful, because seeing the default clearly is how you decide to leave it. An assistant that drafts the tenth empty-state message, so I can focus on the one that appears at a moment of clinical consequence, is not a threat to craft. It removes the friction between me and the decisions that actually need a human.

The pattern underneath all of these: AI is very good at producing volume for the designer to judge. The judgment stays with you.

Where the erosion starts

The trouble begins when generated output stops being raw material and starts being the deliverable. An AI-generated hero illustration on a designer's portfolio. Fake user photography on a product that sells trust. A case study padded with plausible-sounding process that never happened. Each of these works until the moment a viewer recognises it, and designers recognise it fast, at which point it does not just fail on its own terms. It changes how you read everything around it. If the illustration is generated, what else is? That question, once asked, never fully goes away.

This is why I keep AI-generated imagery off my own portfolio entirely. Not out of purity, but out of simple arithmetic. The one thing a portfolio sells is that the work is real and the judgment is mine. Any pixel that undermines that claim costs more than it saves, at any volume.

The line, stated plainly

Here is the rule I actually use, and it fits in a sentence: AI can produce anything I will judge, and nothing I would be embarrassed to defend line by line. Research synthesis, drafts, variants, scaffolding, code for my own tools: all fine. Final imagery presented as craft, testimony presented as human, process presented as mine: not fine.

The teams getting this right are not the ones with the strictest policies or the loudest enthusiasm. They are the ones who made the distinction clear, so nobody has to argue it again on every project.

Takeaways

If you lead a design team, write your own version of the line and publish it internally. Ambiguity is where trust erodes. If you are an individual designer, run the recognition test before you ship anything generated: what happens to your credibility the moment a sharp viewer spots it? And if you are hiring designers, ask them where their line is. The ones who have an answer have thought about their craft. The ones who do not will let the default decide.

The machine is a superb process tool. The artifact still has to be yours, because your name is the part that cannot be generated.