# The End of the Static Screen: Architecting Intent-Driven UX — Gus Iwanaga, commercetools

Source: [AI Engineer](https://www.youtube.com/watch?v=QrMcNe2jjt8)  
Feed7 permalink: https://feed7.dev/p/the-end-of-the-static-screen-architecting-intent-driven-ux-gus-iwanaga-c-0hulcxp  
Published: 2026-09-01T20:00:01.000Z  
Trust: Source Linked (source_linked)

## Why Included

Generative UI needs more than a component catalog: agents require explicit layout rules, schemas, and design-system constraints to produce stable interfaces across repeated requests.

## Source Summary

The same sales-report prompt produced **four different layouts**, with changing labels, KPI cards, charts, and information hierarchy. The proposed declarative path maps intent and tool data to eligible catalog components, then emits a UI specification rendered as native React components.

## Practical Implication

Treat the component catalog as an agent-facing contract. Define schemas, layout slots, composition rules, interaction patterns, and test queries so the model chooses within a curated design space instead of inventing each screen.

## Agent-Ready Context

The same sales-report prompt produced **four different layouts**, with changing labels, KPI cards, charts, and information hierarchy. The proposed declarative path maps intent and tool data to eligible catalog components, then emits a UI specification rendered as native React components.

Treat the component catalog as an agent-facing contract. Define schemas, layout slots, composition rules, interaction patterns, and test queries so the model chooses within a curated design space instead of inventing each screen.

Declarative rendering preserves the design system but does not make composition deterministic. The model can still arrange eligible components poorly, and the talk offers lessons from an experimental product rather than evidence that the approach works reliably at scale.

## Connected Context

Feed7 judgment across 669 accumulated Signals:

This moves agent-directed UI from style prompting into a runtime architecture: intent is translated into a constrained specification backed by native components. It confirms that persistent design rules can bound what agents produce, while the four divergent layouts show that catalog eligibility alone does not ensure coherent composition; evaluation and iteration remain necessary.

- [VoltAgent/awesome-design-md](https://feed7.dev/p/awesome-design-md-0j28iaw) — DESIGN.md supplies persistent tokens and guardrails; the component catalog turns comparable constraints into executable schemas, slots, and composition rules at render time.
- [Ending AI Slop — Thais Castello Branco, Taste Labs](https://feed7.dev/p/ending-ai-slop-thais-castello-branco-taste-labs-1bcbp7n) — Its mixed evaluation approach addresses the remaining gap: deterministic component constraints can enforce alignment and typography, while overall composition still requires preference-sensitive judgment.
- [The Missing Layer: Design Taste in AI Agents — Hassan El Mghari, Together AI](https://feed7.dev/p/the-missing-layer-design-taste-in-ai-agents-hassan-el-mghari-together-ai-00err2g) — The observed layout variation reinforces treating generated UI as a draft and iterating, even when the agent is restricted to approved components.
- [MCP Apps: Extending the Frontier — Ido Salomon & Liad Yosef](https://feed7.dev/p/mcp-apps-extending-the-frontier-ido-salomon-liad-yosef-1gtlsr9) — Both make interactive UI an agent output, but MCP Apps package server-provided views inside a host sandbox whereas this approach has the model compose a native React specification from a local catalog.

## Context Map

- Layer: craft
- Domains: coding
- Topics: design-engineering, interface-quality, tool-use

## Uncertainty

- Declarative rendering preserves the design system but does not make composition deterministic. The model can still arrange eligible components poorly, and the talk offers lessons from an experimental product rather than evidence that the approach works reliably at scale.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.