When visual style stops being a moat
By Allan Leone on
For a decade a recognisable look was something clients paid a premium for and competitors needed months to copy. That gap closed. What is left to sell is the part that was always harder to put in a deck.
A distinctive visual style used to function as a moat. You developed one over years, clients paid for the association, and a competitor who wanted the same effect needed to hire similar people and spend similar time.
That gap is now a few prompts wide. You can hand a model three reference images and get something in the neighbourhood immediately.
What that actually kills
It is worth being precise, because the panicked version of this argument overstates it. What became cheap is imitation of a finished look. What did not become cheap is arriving at a look that is right for a specific problem in the first place.
The business model that dies is the one where the studio's aesthetic was the product and every client got a version of it. That was never really design work. It was execution scarcity wearing a portfolio.
Why models drift toward the middle
A generative model is trained on the average of what already exists, and it is optimised to produce plausible output. Plausible is the operative word. It reliably lands near the centre of its training distribution, which is exactly where the competent and forgettable version of any design sits.
That is not a temporary limitation to be fixed by a better model. It is what the objective function asks for. The deliberately odd choice, the one that reads as wrong until you understand the constraint behind it, is by definition not the most probable next token.
What clients are actually buying now
In practice the thing that survives is the reasoning, which means the deliverable has to change shape. Not just the screens, but the record of why they are like that.
- Why this flow has four steps when the competitor does it in one, and what breaks if you compress it.
- Which feature we argued against, what the client wanted it for, and what we proposed instead.
- Why the empty state got more attention than the dashboard, and what the data said about where people actually arrive.
None of that is copyable in four prompts, because none of it is visible in the output. A competitor can clone the interface and still not know which parts were load-bearing.
Where to start
- Add a decisions section to your next case study. Three decisions, what the alternative was, why you chose. Keep it to a page.
- In the next pitch, lead with the diagnosis rather than the visual direction. If the client cannot tell your diagnosis apart from a competitor's, that is the real problem to solve.
- Audit your own portfolio for pieces where you cannot remember why anything is the way it is. Those are the ones a model can now produce.
- Charge separately for the thinking. If it is bundled into the price of screens, it will keep being valued like screens.
Tags: design, ai, craft