Traditional AI focuses on optimizing decisions within systems. It analyzes data, recognizes patterns, and produces predictions or recommendations, but largely operates behind the interface.
As Nielsen Norman Group explains, traditional interfaces still require users to specify how tasks should be completed through explicit controls and steps. AI improves system intelligence, but the interaction model itself remains unchanged.
Timeline segment highlighting - AI made software smarter, but not easier to interact with. Users still needed to understand interfaces, flows, and controls in order to get value from the system.
Generative AI enables systems to understand intent and generate outcomes—including text, content, actions, and interfaces—in real time.
Nielsen Norman Group describes this shift as a move from users telling systems how to do something to telling them what they want to achieve. The system interprets the goal and determines the appropriate response or interface.
Generative interfaces were preferred by users over conventional interfaces in ~70% of evaluated tasks, showing that adaptive interaction models outperform static or conversational interfaces.
76% of FinTech apps using AI personalization improved user engagement and daily active users by ~41% vs ~17% for limited AI integration.
AI enables several new interaction models that move beyond traditional screens and workflows. These paradigms often coexist rather than replace each other.
| Interaction Model | What It Looks Like | Why It Matters (Reality Check) | Real Product Examples |
|---|---|---|---|
| Zero-UI (Interface-Light) | ・It refers to interaction models where traditional screens and controls fade into the background. | ||
| ・Users interact through voice, gestures, or contextual signals instead of visible interfaces. | Removes friction where screens are impractical, but failures feel intrusive because users didn’t explicitly act | Amazon Alexa routines, Google Nest, Tesla Autopilot, Apple Watch health detection | |
| Multimodal Interaction | ・This allows users to communicate using multiple input methods at the same time—such as voice, touch, gesture, or vision. | ||
| ・The system combines these signals to better understand intent. | Feels natural and fast; poorly designed multimodality creates ambiguity around intent | ChatGPT (image + text), Apple Vision Pro, Google Maps (voice + map), Microsoft Copilot with files | |
| Mixed-Initiative Systems | ・In mixed-initiative systems, both the user and the AI can take the lead. | ||
| ・The system does not just wait for commands—it can suggest actions, ask questions, or intervene when needed. | Boosts efficiency without removing human authority—ideal for professional tools | Google Calendar suggestions, Grammarly, GitHub Copilot, Figma smart guides | |
| Outcome-Based Interaction | ・It focus on completing a goal rather than guiding the user through steps. | ||
| ・The user describes the desired result, and the system figures out how to achieve it. | Eliminates workflow learning but raises stakes around correctness and trust | ChatGPT, Notion AI, Canva Magic Design, Excel Copilot | |
| Agentic Interfaces | ・It allow AI agents to operate semi-autonomously. | ||
| ・Users set goals and constraints, while the agent plans and executes tasks over time. | Scales human capacity but demands transparency, auditability, and strong guardrails | AutoGPT, OpenAI Agents, Salesforce Einstein Copilot, Adept ACT-1 | |
| Predictive Interaction | ・It anticipate what users may need next and surface options before being asked. | Saves time; wrong predictions quickly erode trust and feel invasive | Netflix & Spotify recommendations, Gmail Smart Compose, Figma auto-layout |
Traditional software workflows were built to be linear, explicit, and screen-driven. The interaction logic assumed that users would move step by step through predefined paths.
What traditional workflows looked like: