AI in Design · 6 min read
Using AI in design without losing the craft.
AI in design is neither the apocalypse nor the saviour. It's a new tool, with very specific places where it earns its place — and very specific places where it quietly degrades the work. Here's how we use it at the studio, and the rules we don't break.
Where AI helps the craft.
Three places, consistently. First, exploration: generating mood, palette, and composition variants in the first hour of a project, when the goal is range, not decisions. Second, copy iteration: rewriting a headline 30 ways to find the one with the right rhythm. Third, production: removing backgrounds, generating placeholder imagery, automating asset resizing. None of these is the deliverable. All of them are steps that used to consume hours and now don't.
Where AI quietly degrades the work.
Three places, also consistently. Final hero imagery — generated images at hero scale almost always read as generated, even when they're technically good. Brand voice copy — AI defaults to fluency, not specificity, and a brand's voice lives in the specificity. Identity systems — AI is great at producing one beautiful thing, terrible at producing a system that holds together across twenty surfaces.
The pattern is consistent: AI is excellent at one-shot artefacts, and weak at coherent systems. Branding is mostly the second.
The rule we use: AI before the line, never after.
A simple frame. There's a moment in every project when a decision moves from "exploration" to "this is the answer." We let AI run wild before that line. After it, everything is hand-crafted. AI doesn't get to make final type choices. It doesn't get to write the brand line that appears on packaging. It doesn't render the hero image that defines a brand for two years.
The other rule: AI doesn't get to skip the brief.
A frequent failure pattern: prompt the model, get a beautiful artefact, and try to retrofit a brief around it. The output looks good. It also doesn't say anything specific about the company commissioning it. We never start a project at the model. We start at the brief. The model joins later, when we know what we're trying to say.
What buyers can actually feel.
A common founder worry: will my customers know if I used AI? Honestly, increasingly yes. Generic generated photography reads as generic. AI-typeset slides read as AI-typeset. The audience has spent two years looking at this stuff. Their eye is better-trained than founders give them credit for. The brands that read as premium are the ones whose visible work was clearly considered by humans.
Where AI changes the studio's economics.
AI hasn't made design cheaper. It has made the lower 30% of design work effectively free, which has raised the bar for what counts as professional. The same fee that paid for "competent" two years ago now needs to deliver "considered." The client noticing the difference is faster than ever.
A practical workflow.
On a typical brand identity project, AI shows up in three explicit places. Day one, for divergent mood and reference exploration — fast, broad, deletable. Mid-project, for copy iteration — generating, never finalising. End-stage, for production efficiency — asset variants, format conversions, social cutdowns. The strategy, the type system, the brand voice, the hero work — those stay human. That's where the brand actually lives.
A short test.
If a competitor with the same prompt could produce 80% of your asset, you have a generic asset, not a brand asset. Brand work is the work that prompt couldn't have produced. That's the work to defend.
If you'd like a candid view on where AI fits — and doesn't — in your brand or website workflow, our 30-minute consultation is free.