Designing with AI: Setting the Intent

The designer doesn't disappear, they change tools

Let's clear up the classic fear first: no, AI doesn't replace the designer. It replaces part of the execution work — producing variants, writing the CSS for a shadow, deriving a palette — but it replaces neither taste, nor context, nor responsibility for the result. When Studio Mango's client looks at the landing page tomorrow morning, it's you they'll judge, not the model.

This shift resembles what photography went through with digital: the camera handles the focus, but it's still the photographer who chooses the frame, the light and the moment. With AI, your craft moves upstream (formulating a clear direction) and downstream (sorting, critiquing, refining). The middle — raw production — speeds up tenfold.

Direct consequence: the quality of what you get depends almost entirely on the quality of what you formulate. A fuzzy brief produces a fuzzy result. A precise, embodied brief produces a proposal that surprises you in a good way. This whole chapter is about that skill: setting the intent.

Intent, not recipe

There are two ways to talk to a design AI. The first is the recipe: "put a blue #3B82F6 button, 44px tall, with an 8px border-radius". It works, but you're only using the AI as a CSS typewriter — you're doing all the design work yourself, and you're depriving yourself of its proposals.

The second is the intent: "a CTA that inspires trust for a wellness app, understated, premium, calming". There, you describe the emotion sought and the use, and you let the model translate that into visual decisions. It might propose a deep sage green with generous type — something you wouldn't have specified, but that you recognize as right. The AI proposes, you choose. You remain the art director.

The good practice is to mix the two levels: intent for the overall direction, and precise constraints where you've already decided (the brand color imposed by the client, a licensed typeface, an image format). Everything that isn't constrained becomes a space for proposals. It's exactly like briefing a talented junior designer: you frame the why, you leave latitude on the how.

Learning to name what you see

To formulate an intent, you need vocabulary. It's the hidden skill of designing with AI: the more you can name things, the more effective your prompts. "I want it to breathe" becomes "increase the vertical spacing between sections to at least 96px". "It's cluttered" becomes "the visual hierarchy is confused: the title, the subtitle and the CTA are competing for attention".

A few notions to master absolutely, because they come up in almost every design exchange: visual hierarchy (what does the eye see first, second, third?), contrast (of color, size, weight — it's what creates hierarchy), vertical rhythm (the regularity of spacing that makes a page calm or chaotic), and density (the amount of information per screen). When you critique a render with these words, the AI understands exactly what to fix.

Good news: the AI can help you acquire this vocabulary. Show it an interface you admire and ask "describe what works in this design in terms of hierarchy, contrast, space and typography". You get a reasoned analysis — and the words to formulate your next briefs.

References are worth a thousand words

Describing a style from scratch is hard, even for an experienced designer. References short-circuit the problem: instead of abstractly explaining "premium and calming", you cite products whose visual DNA is well known. The AI knows the major art directions of famous products — Apple, Stripe, Linear, Notion, Headspace — and what characterizes them.

The important subtlety: never cite a reference without specifying what you're taking from it. "Like Apple" is ambiguous — the generosity of space? the product photography? the tone? Be surgical: "Apple's generosity of space, Headspace's soft color palette, Linear's crisp typography". Three precise signals that the model fuses into one coherent direction, rather than a servile imitation of a single product.

Art direction for a meditation app:
- Apple's generosity of space
- Headspace's soft color palette
- Linear's crisp typography
AVOID: garish gradients, clutter, the "generic AI template" look.

Rephrase this direction in 3 sentences that will serve as a reference for the rest of the project, then propose 3 adjectives that sum up the mood.

Saying what must be avoided

Here's the most underrated lever of the brief: the list of don'ts. AI models have statistical habits — visual choices they produce by default because they're over-represented in their data: purple-blue gradients, emojis in headings, cards with identical soft shadows. If you don't explicitly exclude them, you'll get them.

An "avoid" list acts as an extremely powerful negative filter. "No garish gradients, no emojis, no clutter" rules out 80% of generic renders right away. It's often more effective than piling up positive adjectives, because don'ts are binary and verifiable: either there's an emoji in the heading, or there isn't.

Building the art direction brief

Let's put it all together. A good art direction brief fits in five blocks: the context (which product, which audience, which business goal), the emotion sought (three adjectives maximum, otherwise everything blurs), the references (two or three, with what you take from each), the hard constraints (brand color, accessibility, format) and the don'ts. Five blocks, about ten lines — it's short, and that's deliberate: a one-page brief nobody reads is worth less than a ten-line brief everyone remembers.

flowchart TD
  C["Context: product, audience, goal"] --> E["Emotion: 3 adjectives max"]
  E --> R["References: 2-3 products + what you take from each"]
  R --> K["Hard constraints: brand, accessibility"]
  K --> I["Don'ts: the anti-generic list"]
  I --> B["Art direction brief"]
  B --> P["Every prompt in the project"]

CONTEXT: landing page for Sereno, a meditation app for stressed urban professionals. Goal: conversion to the free trial.
EMOTION: calm, premium, confident.
REFERENCES: Apple's generosity of space, Headspace's soft color palette, Linear's crisp typography.
CONSTRAINTS: AA contrast minimum, mobile-first, no stock photos.
DON'TS: purple gradients, emojis in headings, generic soft shadows, clutter.

Rephrase this brief as a 3-sentence art direction, then list 5 concrete visual decisions that follow from it.

The human's role: sort, arbitrate, own it

The AI generates fast, but it generates everything with the same confidence: the brilliant and the mediocre alike. Your job is to sort. Faced with three proposals, ask yourself: which one best serves the goal (here, conversion to the free trial)? Which one is consistent with the brand? Which one will survive six months of product evolution? That judgment cannot be delegated.

There's also a question of responsibility. If the generated palette has insufficient contrast and visually impaired users can't read the page, it's not "the AI's fault" — it's yours. If the landing page looks like three competitors because you accepted the first render without challenging it, same thing. The AI is a partner, not an alibi. That's what separates a memorable design from a patchwork of pretty ideas: someone, somewhere, owned the choices.