Design Is Not the Last 10%
Treating design as decoration applied at the end is expensive- not because the result looks worse, but because by then every difficult decision has already been made.
There is a version of the software process where design arrives near the end. Requirements are agreed, architecture is settled, development is underway, and a designer is brought in to make the screens presentable.
The output of that process is usually fine. That is precisely the problem- fine is what you get when the interesting decisions were made months earlier by people optimising for something else.
Design decides what gets built
The instinct is to treat design as being about appearance. In practice, the highest-value design work happens before any visual choice is made, and it is about scope.
Sketch a workflow and you immediately discover which steps exist because someone genuinely needs them and which exist because the database has a table. You find the screen that three different roles need to see differently. You find the field everyone assumed was required that nobody can justify. You find the two features that are really one feature described twice.
Those discoveries are the expensive ones, and their cost scales brutally with when you make them. Deleting a step in a wireframe costs a pencil. Deleting the same step after it has been built, tested, documented, and integrated costs a sprint and an argument.
Prototypes settle disputes that documents extend
Written requirements are read differently by everyone who reads them. Two stakeholders can agree enthusiastically on a paragraph while holding incompatible pictures in their heads- and neither discovers this until something is demonstrated.
A clickable prototype ends that ambiguity in an afternoon. Nobody argues about interpretation while looking at the same screen; they point at it and say what is wrong. This is not a presentation technique. It is the cheapest error-detection mechanism available in software.
Consistency is a system, not a habit
As a product grows, small inconsistencies compound: four shades of the same blue, three button styles, two different date formats, a modal that closes on outside-click in one place and not another.
Individually, none of it matters. Collectively it produces a product that feels unreliable- users cannot form a stable model of how it behaves, so they proceed cautiously, which reads as difficulty.
A design system is the fix, and it is best understood as engineering infrastructure. Defined tokens, a fixed component set, documented rules for when each is used. Its value is not aesthetic consistency; it is that a developer building screen forty makes the same decisions as the developer who built screen three, without needing to ask.
The full range still matters
None of this diminishes the visible craft. Brand identity, illustration, motion, 3D, packaging, the social graphics- these carry your reputation into places you never get to explain yourself. A well-made brand does real commercial work.
But that work lands harder when it sits on top of a product whose structure was designed rather than inherited. Polish applied to a confusing flow produces a beautiful confusing flow.
Bring design in when the questions are still open. That is when it is cheapest and when it is worth the most.
