Prompt engineering for design means writing structured instructions (role, context, objective, constraints, style, output format, quality criteria) so AI models produce reliable, reusable results inside a product design workflow: research, ideation, design systems, interface generation, content and visual production.
A prompt is an interface
I treat a prompt like a screen: it has a user, an objective, a structure, error states. A good system prompt reads like a spec.
ROLE Senior product designer + UX strategist
CONTEXT B2B SaaS · data audience · premium brand
OBJECTIVE Clear, desirable, measurable onboarding
CONSTRAINTS AA accessibility · mobile-first · existing design system (tokens v3)
STYLE editorial · high-contrast · restrained motion
OUTPUT flow + annotated wireframes + UX copy
QUALITY hierarchy readable in under 3 s · 1 action per screen
This seven-block structure is the base of all my prompt systems. It is detailed step by step in Prompt engineering for designers: the method.
Seven areas of application
| Area | What AI does well | What stays with the designer |
|---|---|---|
| UX research | Interview synthesis, clustering quotes, spotting themes | The interview guide, interpretation, decisions |
| Ideation | Flow variants, structural alternatives, counter-proposals | The choice and its rationale |
| Design systems | Component documentation, consistent naming, spotting inconsistencies | The foundations, governance |
| UI generation | First drafts of screens in code, states, responsive variants | Hierarchy, finish, consistency |
| Content | Microcopy, error messages, help text, multilingual versions | Tone, truth, legal review |
| Visual production | Campaign visuals, illustrations, format variations | Art direction, selection, retouching |
| Workflows | Automation scripts, quality checks, QA | Defining what a good result is |
What I deliver
- A prompt system: templates per use case, versioned, with sample inputs and expected outputs.
- Tooled workflows: documented chains (for example: interviews, transcription, synthesis, personas, user flows), with human checkpoints.
- Design agents: reusable configurations for specific tasks, for example an agent that checks design system consistency or generates a component's missing states.
- Short training: half a day so the team owns the system and knows how to extend it.
- Usage guidelines: what we hand to AI, what we do not, and how we check.
Observed results
On my own products, AI changed the speed of certain tasks without lowering the quality bar:
- Synthesis of a ten-interview study: from one week to one day, with a full reread of the quotes.
- First version of a realistic front-end prototype: from three days to a few hours, then manual finishing.
- Documentation of a forty-component design system: written in two days instead of two weeks, from the real components.
- Visuals for a campaign of twelve client demos: no photo budget, with consistency held by a locked prompt system.
These figures are orders of magnitude from real projects, not promises. What does not speed up: understanding the problem, making trade-offs and owning the result.
Who it is for
- Design and product teams that want to scale their use of AI without losing quality.
- Agencies and studios looking to produce more variants and adaptations on the same budget.
- Founders who want to prototype fast and clean before hiring.
- Innovation teams that need a credible usage framework and concrete examples.
How it connects to my other services
Prompt engineering does not replace UX strategy: it serves it. It does not replace art direction: it equips it.

