· 4 min read
A template can say no
A useful design template records the audience, decision, hierarchy, and vetoes that make a polished output fit its task.
Ask a model for a clean, modern analytics dashboard and you will get one, in roughly the sense that asking a kitchen for "something nice" gets you food. Now compare that brief with one that names the person using the page, the decision she is making, the facts that must stay visible together, and three reasons to throw away an otherwise handsome result. Both leave room for visual invention. Only the second can tell you whether the invention worked.
That second brief is where taste actually lives when you hand design to a template. Not in the token for "modern", which carries no audience, no hierarchy, and no stopping rule — in the priorities and the refusals.
The same task permits different pages#
The open brief makes familiar dashboard furniture easy to justify. Backlog gets a card, staffing gets a card, breach risk gets a third, the queue table slides below a chart. Every piece is plausible on its own, and together they hand the support lead an inventory when she came for a decision: which queue needs a person moved, right now?
| Open brief | Brief with vetoes | |
|---|---|---|
| Brief | “Make a clean, modern analytics dashboard” | Names the support lead, staffing decision, required facts, inherited table, and forbidden failure |
| Composition | Metric cards, chart, table, and several equally prominent controls | Breach risk and staffing gap stay adjacent, with the queue table as the main body |
| Decision support | Backlog is visible, but the staffing decision has no privileged path | One move-agent action is evaluated against the facts that justify it |
| Reason to reject | Only a general judgment that the page does not feel right | Reject if records disappear, actions compete, or status depends on color |
Under the constrained brief, breach risk sits beside the staffing gap because those two facts jointly answer her question. The table survives because a summary may not replace the records an operator has to inspect, and status wears a label as well as a color. The page can still turn out ugly — but its important failures now have names a reviewer can point at, which is a real improvement over "it feels off".
I should say plainly that this is a constructed comparison, not a benchmark. The claim it supports is narrower and, as far as I can tell, solid: changing a few lines of brief changes which design choices can be defended in review. Creativity research has a name for the half that "modern" skips — Runco and Jaeger's standard definition pairs originality with effectiveness, and effectiveness means fit: this institution, this decision, this reader.
Write the refusal into the brief#
Watch what each added line buys.
- Make a clean, modern analytics dashboard.+ Audience: support lead deciding which queue needs staff.+ Keep backlog, breach risk, and staffing gap visible together.+ Preserve the existing table; summaries may not replace records.+ One primary action: move an agent between queues.+ Use the product type scale and spacing tokens.+ Reject any layout where color carries status alone.The audience line fixes a point of view. Keeping three facts together defines the page's central relationship, and preserving the table stops a common "simplification" from deleting the evidence. One primary action settles the fight between buttons. The last line catches the page that looks perfectly coherent to a sighted reviewer while hiding status from anyone who cannot split its colors — the failure a polished render is best at smuggling through.
UICoder ran this principle at training scale: generate interface programs, compile them, render them, score the renders, drop the near-duplicates. The paper reports that 0.4 percent of the generated examples survived. Generation made the options cheap; the filters that discarded the other 99.6 percent were the taste.
A template can preserve the wrong taste#
Vetoes are inherited decisions, so they age like any other decision. A template built for a calm marketing dashboard will quietly suppress the density an incident console needs, and a mandatory three-card summary can outlive the product it summarized. Reuse makes decisions consistent, including the stale ones — the refusals need review too, on some schedule nobody has invented yet.
For the support dashboard, that review at least has material to work with. Breach risk drifted away from staffing, the records vanished, status leans on color alone: specific defects, each traceable to a line in the brief. The sharpest of those lines is still the last one, and it is doing more design work than the word "modern" ever did.