FEI HUANG
I build the infrastructure that designers and engineers
rely on.
PRODUCT DESIGNER I SYSTEMS • ENTERPRISE WORKFLOWS
From design infrastructure to AI behavior architecture
PREVIOUSLY BUILT SYSTEMS AT
FEI HUANG
FEI HUANG
Meta · Evolving one design system · 03 of 03
AI Interaction Architecture
Foundations for AI-Assisted Workflows
This work reflects early-stage design system exploration as AI capabilities were introduced into product workflows
Scope & Role
Role: Lead Product Designer
Scope: Platform level, cross-product
Partners: Design Systems, Product Design, Platform Engineering
System AI Interaction Model
synthesized from exploration work
Context
In early 2025, generative AI capabilities were rapidly integrated into business workflows, and products started shifting from configuration-driven interfaces to goal-oriented, AI-assisted systems. At that time, however, there was no shared design system model for how AI should behave across products.
This work marks an early exploration of a design system focused on defining AI behaviors, interaction constraints, and system states to help guide teams as they begin to experiment.
Project Overview
As AI-assisted features started appearing on different surfaces, such as suggestions, generation, and recommendations, product teams were solving similar interaction problems independently. This was especially clear in workflows like automated campaign setup (e.g., Advantage+), where reduced manual configuration increased reliance on AI-driven decisions.
These conditions revealed new challenges regarding trust, control, and system responsibility—issues that current UI patterns did not address.
At that time, there was no consistent guidance on:
• how AI suggestions should be surfaced
• how system states should be communicated
• how user control should be preserved
I led an initiative within the design systems team to identify these emerging patterns and translate them into a more structured interaction model, rooted in real product explorations and early prototypes.
The goal was not to deliver a fully production-ready system but to create a shared foundation that teams could align around as AI usage grew.
System Gaps
Through collaboration with design and engineering partners, we identified recurring gaps across teams, including:
• Inconsistent approaches to presenting AI suggestions and partial results
• No shared model for communicating confidence levels or uncertainty
• Ambiguity regarding when AI should act versus when it should defer to users
• Lack of consistency in handling failure, fallback, and recovery
These gaps risk fragmentation and inconsistent user experiences as AI capabilities grow.
What I Shaped
This work focused on structuring early patterns into a system-level interaction framework to guide how AI behaviors could be consistently integrated across workflows.
Key contributions included:
• Mapping AI system states and transitions (e.g., idle → processing → completed/failed) to establish a shared model for system behavior
• Defining core interaction patterns for AI-assisted workflows, including suggestion, confirmation, and override, to clarify how AI proposes, acts, and defers to users
• Establishing initial guardrails around user control, risk levels, and system responsibility to promote safer decision-making in high-impact workflows
• Creating component-level explorations and prototypes to validate interaction flows in collaboration with engineering
• Developing design-to-engineering alignment artifacts (states, props, annotations) to enable consistent interpretation and implementation discussions
The artifacts shown represent a structured synthesis of these explorations, illustrating how these interaction patterns could scale within a design system.
AI system states
In product examples
Banner & Decision log
Impact
Immediate impact within the design system team
• Created a shared language for discussing AI interaction behavior.
• Helped align design and engineering on key system concepts (state, confidence, control).
Near-term impact
• Reduced the risk of fragmented approaches as more teams began building AI features.
• Provided early direction for how AI interactions could be structured consistently.
Longer-term organization-wide foundational impact
• Established early thinking that informed how AI interaction patterns could evolve into a more scalable system over time.