When was the last time you actually critiqued the interface of your AI tools? Not the images they spit out, not the copy they generate — the thing you're actually staring at and clicking on for hours a day. The prompt box. The settings panel. The regenerate button you hammer like a slot machine lever.
We spend so much energy debating whether AI output is good enough. But there's a more immediate problem hiding in plain sight: the tools themselves are, by any design standard, pretty badly made.
What happened
Two things caught my eye this week that, at first glance, seem unrelated.
First, there's a sharp piece doing the rounds arguing that system tools — the functional, unglamorous interfaces people interact with out of necessity — deserve the same design care as any consumer-facing product. The core argument is simple: when a tool can't be made invisible, when people have to see it and use it directly, you can't dismiss its experience as unimportant. Function shapes form, sure. But when function is front and centre, form matters more, not less.
Second, Hoefler & Co. released Gotham Variable — a proper variable font version of one of the world's most recognisable typefaces. Instead of choosing from a fixed set of weights and widths, designers now get continuous control along a spectrum. One file. Infinite precision. You dial in exactly what you need.
So what connects a UX philosophy piece about system tools to a font release? More than you'd think.
The bigger picture
AI tools have become system tools for designers. They've embedded themselves in our daily workflows — generating concepts, iterating on layouts, writing copy, extending backgrounds. And like most system tools, their interfaces are an afterthought.
Think about it. Most AI generation tools give you a text box and a button. Maybe a dropdown for "style" with labels like "cinematic" or "illustration" that mean different things depending on the phase of the moon. You type, you generate, you get something back. If it's wrong — and it usually is, at least partially — you regenerate. Or you rewrite your prompt. Or you start over.
This is not a well-designed interaction. It's a guess-and-check loop dressed up as a creative tool.
The system tools argument nails exactly why this matters. AI isn't invisible in our process. We're not pressing a button and walking away. We're sitting in front of these interfaces, wrestling with them, trying to coerce them into understanding what we actually want. That means the experience of using them is a design problem — and right now, almost nobody is treating it like one.
Compare this with what Gotham Variable represents. Here's a tool that gives designers continuous control. Not "pick from these 12 options." Not "type what you want and hope for the best." A smooth spectrum where you can land precisely where you need to be. That's what good tool design looks like — it respects the designer's judgement and gives them the means to exercise it.
Most AI tools do the opposite. They strip away control and replace it with randomness. You don't adjust — you regenerate. You don't refine — you re-prompt. The interface assumes that more output compensates for less precision.
It doesn't.
Tool spotlight
If you work with type at all — and let's be honest, that's most of us — Gotham Variable is worth a proper look. Variable fonts have been around for a while, but having a typeface this widely used and this well-drawn available in variable format is a genuine workflow improvement.
The practical benefit is twofold. For web work, you're replacing multiple font files with a single one, which means faster load times. For design work, you're no longer locked into predetermined weights. Need something between Medium and Bold? You've got it. Want to fine-tune width for a tricky layout? Go ahead.
It's the kind of tool upgrade that doesn't make headlines but quietly makes your work better. And it's a useful reminder of what designers should be demanding from all their tools: precision, flexibility, and control.
So what do we actually do about this?
Start evaluating your AI tools the way you'd evaluate any design tool — by the quality of the interaction, not just the output.
Ask yourself: does this tool let me refine, or just regenerate? Does it give me meaningful controls, or just a text box and vibes? Can I make a small adjustment without starting from scratch?
The AI tools that will actually stick in design workflows long-term won't be the ones that generate the flashiest images. They'll be the ones that feel like well-designed instruments — responsive, precise, and respectful of the person using them.
We're designers. We know what good tools feel like. It's about time we started demanding that standard from AI, too.