Architecture Next 2026: Designers in the loop — The state of AI in UX/UI

This 30-minute talk by Gil Ayalon, CTO at MonkeyTech (CodeValue Group), was recorded at the Architecture Next 2026 Builders Track in June 2026. It gives a grounded picture of how AI is actually being used across the UX and UI design process — not the speculative version, but the operational one that practitioners are navigating now.

Who it is for

Product designers, design leads, and developers who work closely with design. The talk assumes familiarity with standard design workflows and focuses on what is changing in practice rather than what might change in theory. It is useful at a point when AI tools are available but teams are still figuring out where they genuinely help versus where they add noise.

What it covers

Ayalon structures the talk around the design process stages — research, ideation, wireframing, visual design, prototyping, and handoff — and examines what AI tools are doing in each. He draws a distinction between stages where AI has matured enough to accelerate real work and stages where the outputs still require substantial human correction before they are useful.

A central thread is the question of where designer judgment is being displaced versus where it is being redirected. Ayalon argues that AI tools are not reducing the amount of judgment required — they are moving it. Decisions that used to happen late in the process (selecting between two comps, refining copy in context) now need to happen earlier, in how you frame the prompt, what design system constraints you establish, and which reference materials you include.

The handoff section is particularly practical. Ayalon looks at what AI-assisted code generation produces at the Figma-to-code boundary and what developers actually receive. He is direct about the gap: generated code often works but does not always reflect the design system or accessibility requirements without explicit instruction, and catching those gaps before handoff rather than after is now a designer responsibility rather than an engineering one.

Key takeaways

1. AI accelerates early exploration significantly — generating ten layout directions in the time it used to take to sketch three. The value is in the speed of divergent thinking, not in the quality of any single output.

2. The research synthesis step benefits most consistently from AI in mature teams. Summarizing user interview transcripts, clustering themes, and pulling representative quotes from large datasets are all tasks where current tools are reliable enough to save hours per sprint.

3. Visual design generation remains uneven. Output improves substantially when the model has access to a real design system rather than generating styles from scratch. The constraint is an asset.

4. Prototype fidelity from text-to-UI tools has improved enough in 2026 that some teams use AI-generated prototypes for early usability testing, but the consensus is that you still need a designer to review for interaction logic before showing them to participants.

5. Handoff quality depends on how explicitly you specify accessibility, component naming, and token usage upfront. These are not things AI infers well from visual design alone.

Worth watching if

You are trying to decide which stages of your design process to invest in tooling, or you are leading a team that has adopted AI tools unevenly and wants a structured way to evaluate where the friction actually is. The talk is less useful if you are looking for tool comparisons — Ayalon deliberately avoids naming specific products to keep the analysis durable.