Claude Design is not a shortcut to taste
A practical note on using AI design tools for exploration without outsourcing judgement, taste, or the final call.
Claude Design is interesting because it changes the cost of trying a direction.
That is not the same as changing the value of the direction.
The useful bit is speed at the sketch layer. You can describe a landing page, a slide, a one-pager, or a prototype and get something back quickly enough to react to it while the thought is still warm. That matters. A lot of design time is lost before the work is even good enough to criticise.
But a cheaper first version does not remove the need for taste. It just gives taste more material to work on.
The first version is a provocation
I do not trust first drafts, whether they come from a person or a model. A first draft is not there to be admired. It is there to reveal what the brief forgot to say.
An AI design tool can produce a plausible layout very quickly. The danger is that plausible can feel finished. It has spacing. It has a hierarchy. It has a nice soft gradient because apparently the internet now runs on nice soft gradients. If everyone is tired, that can look like enough.
It usually is not enough.
A good design direction has to answer more awkward questions. What is this page actually trying to make clear? What does the user need before they believe the call to action? Which part of the visual system is doing useful work, and which part is just standing there holding a glass at the party?
Claude Design can help you reach those questions faster. It cannot answer them on your behalf.
Exploration is not judgement
Exploration gets cheaper with AI. That is genuinely useful.
I like being able to test a few tonal directions before committing: quieter, sharper, more technical, more editorial, more visual, more restrained. For client work, that can make the early conversation better because everyone is reacting to something concrete rather than arguing over a sentence in a brief.
But judgement still happens in the edit.
The edit is where you decide that the clever interaction is too clever, the headline is saying the same thing twice, the visual metaphor is fighting the product, or the whole thing looks impressive but does not help the person understand what they came for.
That is not a small detail. That is the job.
Brand systems become more important
AI design tools make it easier to produce more options. They also make it easier to produce options that quietly drift.
One minute the work has a strong visual rhythm. Three prompts later it has borrowed the typography of a SaaS dashboard, the colours of a finance app, and the confidence of a conference keynote. None of those things may be wrong in isolation. Together they can make the work stop sounding like itself.
This is where a brand system earns its keep. Not as a rigid decoration manual, but as a set of constraints that helps you say: this belongs, this does not, this is close but needs tuning.
A good system gives the model less room to average everything out. More importantly, it gives the person reviewing the work a better basis for saying no.
The work still needs a point of view
The more capable the tooling gets, the more valuable the point of view becomes.
If a tool can produce ten directions, the hard part is knowing which one deserves a second pass. If it can turn a rough idea into a prototype, the hard part is knowing whether the prototype explains the product. If it can make something look polished, the hard part is knowing whether the polish is hiding a weak idea.
That is design work. It has always been design work.
Claude Design is not a replacement for taste. It is a faster way to put taste under pressure.
Used well, that is a useful thing. Used lazily, it is just a nicer-looking shortcut to the average.
AI didn’t replace me - it made the case for me is the larger version of the same argument: the tools have changed, but the judgement is still the scarce part.