A convention becomes a convention by disappearing. Nobody notices the scrollbar, the tab, the underlined link — which is precisely the evidence that they work. The corollary is uncomfortable for designers: any interface element that is being widely discussed is, almost by definition, not yet working.

A great deal is currently being discussed. The conventions established between roughly 1984 and 2010 assumed a pointer, a page and a document, and every one of those assumptions is now optional. What is unclear is which of the replacements are genuine improvements and which are simply new.

Three tests a convention has to pass

Interface historians tend to converge on the same criteria, and they are more demanding than they look. A convention has to be learnable in a single exposure, guessable in the second, and survivable when guessed wrong. The third is the one that kills most proposals. A gesture that does something unrecoverable when misread will never become invisible, however elegant it is.

The measure of an interface is not what it does when you use it correctly. It is what it does when you use it wrongly at four in the morning.

By this standard, much of the last decade's gestural vocabulary has failed quietly. The swipe that archives, the long-press that reorders, the pull that refreshes: all learnable, most guessable, and several of them still catastrophic when triggered by accident. They persist not because they passed the test but because the platforms that shipped them are large enough to make the test irrelevant.

The return of the visible control

There is a quiet counter-movement, and it is being led by teams building for high-stakes or low-attention contexts — medical software, industrial control, anything used while tired. Their conclusion is consistently the same: visible controls, explicit labels, no hidden state, and a strong preference for an interface that looks slightly dated over one that requires a tutorial.

  • Every hidden affordance is a training cost paid by the user rather than the product.
  • Reversibility is worth more than efficiency in any interface used under pressure.
  • A label costs a few pixels; a misread gesture costs the whole interaction.

In ten years some small number of the current proposals will be invisible and the rest will be a curiosity. The reliable prediction is not which ones — nobody has ever been good at that — but that we will not be able to remember having learned the survivors.