Your Design Process Isn't Ready for AI Yet. Here's What's Missing

User Testing: Fix It Before the AI Makes It Permanent

‍

In the Kello project, 11 domain experts validated the prototypes before development began. This has long been standard practice at ZIMA, but in an AI-driven development workflow it becomes even more important.
‍

An AI agent will faithfully implement whatever it receives. Every design issue that slips through user testing has a good chance of ending up directly in production code—potentially affecting the entire product.
‍

Testing before handing designs over for implementation means fewer iterations during development. Updating a Figma file is far less expensive than correcting production code. It's the difference between fixing a typo before sending an email and trying to correct it after it has already reached the entire company.
‍

Two practices become especially important:

• Interactive prototypes used for testing should be polished enough to serve not only as a research tool but also as a visual reference for the AI agent.
• Document research findings as concrete design changes in Figma rather than high-level observations in a presentation. "Users struggled with the form" is an observation. "Move the Date of Birth field above the Phone Number field because users expected it there" is an actionable instruction that an AI agent can implement correctly.

‍

‍

‍

AI as the Most Meticulous QA Reviewer You'll Ever Have

‍

One of the biggest changes introduced by AI agents is that visual QA can now be largely automated. AI agents are capable of comparing the implemented interface against the original Figma design—for example, by matching browser screenshots with design mockups using tools such as Playwright.

‍

This has direct implications for the way we design interfaces.

• Clean, flat layouts are easier for AI agents to verify than highly animated or visually complex interfaces. Simpler structures leave less room for discrepancies between design and implementation.
  ‍
• UI states—hover, active, disabled, error, and others—should be designed as separate component variants rather than hidden layers inside a single component. AI agents compare each state individually against what is rendered in the browser.
  ‍
• Responsive layouts should be prepared as dedicated frames for every breakpoint. This gives AI a clear and unambiguous visual reference for desktop, tablet, and mobile versions instead of relying on hidden variants within a single artboard.
  ‍

In other words, the way a Figma file is structured is no longer just a matter of designer convenience. It directly determines whether automated visual verification can work reliably..

‍
The Designer as the Product Owner of AI-Driven Delivery

‍

‍

You don't need to know how to write code to understand how AI agents implement designs. But understanding that implementation process is becoming a core design skill—just as important as accessibility guidelines or usability heuristics.
‍

In practice, this means adopting a few new habits:

• Test your own designs using tools such as Figma MCP. You'll see exactly what the AI agent "sees" in your file and identify potential issues before a developer comes back with a list of questions.
  ‍
• Talk to developers not only about whether the interface looks right, but also about whether interaction details are described clearly. These conversations uncover documentation gaps much faster than visual feedback alone.
  ‍
• Think of your design as a technical specification with a visual layer—not the other way around. Visual quality still matters, but the precision of your documentation increasingly determines the quality of implementation.
  
  ‍
  ‍

A Process Ready for AI-Driven Delivery Is Still Simply a Good Design Process

‍

‍

User testing before implementation, designing with automated verification in mind, and designers who understand the entire implementation process are not revolutionary ideas. They are the same principles of good design we've always followed—now executed with a much higher level of precision.
‍

At ZIMA UX, UI & Design Strategy, we already design with AI-driven delivery in mind. This isn't a change in visual style—it's a change in how we think about what design should communicate, and who—or what—it needs to communicate with.
‍