UX-First AI Workflow: How I Use AI Without Losing Control Over Design
AI can generate interfaces in seconds. But good UX doesn't come from prompts. My workflow: from systemic thinking to an orchestrated design process.
AI can generate interfaces in seconds today: fast, clean, impressive at first glance. But that’s exactly where the problem lies.
Beautiful surfaces don’t automatically mean good user experience. Real UX doesn’t emerge from a prompt — it comes from context, prioritization, user understanding, and above all, clear decisions.
This article walks through my concrete workflow. Three steps that make the difference between “using AI” and “deliberately steering AI in design.”
From Designer to Orchestrator
The role is changing. Anyone designing digital products today is no longer just responsible for visual interfaces. The value shifts from pure execution toward orchestration.
That means: You’re not just creating screens or variations. You define the rules by which good interfaces can even be created. You set the boundaries within which the AI is allowed to work. You define design systems, components, user flows, and UX rules.
That’s why my workflow doesn’t start with perfect interfaces. It starts with structure — not pixels, but logic, priorities, and the system behind the interface.
“Great digital products aren’t created by AI quickly spitting out beautiful surfaces. They’re created when you give the AI clear guardrails.”
Step 1: Systemic Thinking
Before I build an interface, I clarify:
- What problem needs to be solved?
- What user situation is at the center?
- What content is truly relevant?
- What structure emerges from that?
I don’t think in screens first. I think in information architecture, UX flows, components, states, and rules. That’s precisely the difference between classical execution and orchestration.
Practical approach:
- Sketch the user journey (paper or Miro, no tool hype needed)
- Content inventory: What needs to go on which page?
- Prioritize: What’s essential, what’s nice-to-have?
- Define components: Which building blocks repeat?
- Set rules: How does the system behave in each state?
Only when this structure is in place does it make sense to use AI for visual implementation.
Step 2: Precise Prompting
Precise prompting isn’t about finding the magical one-liner that makes everything perfect. It’s precise design communication.
The quality of AI output depends not only on the model but on how well you can describe:
- What visual direction is desired
- What UX principles apply
- What hierarchies matter
- What constraints the system has
- What should deliberately not happen
Concrete prompt example from my workflow:
“Create a landing page section for a local sharpening service. The headline must explain what the service is in 3 seconds. Below it, three service areas as cards. No generic icons — concrete descriptions. The CTA button leads directly to the contact page. Mobile first.”
That’s not a magic prompt. It’s a precise task description. And that’s exactly the point: Good prompting is good briefing. If you can’t clearly tell your designers (human or AI) what you want, you’ll never get good results.
Step 3: Use AI Confidently as a Tool
For me, tools like ChatGPT, Claude or Gemini are neither a threat nor substitute designers. They are work partners within a system.
I use them specifically for:
- Structuring — Organizing information, suggesting hierarchies
- Ideation — Drafting initial directions and variations
- Reviews — Checking existing interfaces against UX principles
- Faster iteration — Generating, comparing, discarding variations
The decision stays with the human. The AI provides speed, you provide direction — whether in AI automation of workflows or the design of user interfaces. The AI produces variations, you make the decisions.
Conclusion
The new value of design no longer lies solely in building good interfaces as quickly as possible. It lies in thinking about and steering systems in a way that produces better products.
The designer becomes the orchestrator: orchestrating AI tools, understanding user needs, and making quality-driven decisions where AI alone reaches its limits.
The three steps summarized:
- Think in systems, not screens — before you open any tool
- Communicate precisely — good prompting is good briefing
- Use AI as a partner — it provides speed, you provide direction
I covered this whole topic in detail on my YouTube channel:
