Interface
Human-AI Interaction — designing the touchpoint where users actually experience and interact with the value of AI.
Role of this domain
Interface is the only touchpoint in the five-domain framework where users directly experience AI. No matter how strong the infrastructure and orchestration underneath it are, AI adoption fails if the user experience is bad.
flowchart LR
A["⚙️ Orchestration<br/>Internal processing"] --> B["🤝 Interface<br/>User touchpoint"]
B --> C["User<br/>Perceived value"]
B --> D["Feedback<br/>collection"]
D --> A
style B fill:#16A34A,stroke:#15803D,color:#fff
style A fill:#7C3AED,stroke:#6D28D9,color:#fff
style C fill:#EA580C,stroke:#C2410C,color:#fff
style D fill:#2563EB,stroke:#1D4ED8,color:#fff
Core components
| Component | Description |
|---|---|
| UI/UX design | Conversational UI (CUI), agent-run progress and control, multimodal optimization |
| Multimodal input | Handling and visualizing images, voice, diagrams, and other input types |
| AI literacy | Guidance that helps users make effective use of AI |
| Feedback loops | Collecting user feedback and feeding it back into the system |
Core focus: managing psychological acceptance
Interface is not simply a technical UI problem — it’s the domain that manages users’ psychological acceptance of AI.
- How do users react when the AI is wrong?
- Is there over-reliance on, or under-trust of, AI output?
- Can users control how they collaborate with AI themselves?
Agents raise the stakes on all three. A chat response is wrong in one visible place; an agent running a multi-step harness is wrong somewhere inside a sequence of actions that already had effects. The interface obligations shift accordingly — show the plan before it runs, report progress while it does, allow interruption at any point, and require approval before anything irreversible.
Health check questions
“Do users understand the limits of AI and use it appropriately?”
- Does the interface clearly indicate that content is AI-generated?
- Is user feedback (thumbs up/down) actually connected to system improvement?
- Does multimodal input reflect users’ real needs?
- Is AI literacy training provided on a regular basis?
- When an agent runs for more than a few seconds, can the user see what it is doing and stop it?