The First New UI Paradigm
in 60 Years
We are witnessing a fundamental shift from
Command-Based Interaction
(Do this, then click that) to
Intent-Based Interaction
(Here is the result I want). Current apps are static
factories. Future apps are liquid.
But this shift creates a new usability crisis:
The Affordance Gap.
Command Line
High friction. Users had to learn the machine's language to operate software. Total Recall required.
Graphical UI
Static menus and buttons. Users forced to memorize rigid paths. Total Recognition. "Don't Make Me Think."
Smart Apps
Intent-Based: The interface adapts to the user, not the other way around.
The Affordance Gap
An analysis of the "Articulation Barrier" in AI interfaces versus the "Click Fatigue" of traditional hierarchies.
Faster Execution
Speed gain for information retrieval tasks using GenUI vs traditional click paths.
Error Risk
Increase in incorrect outcomes when users over-rely on AI intent without verification.
Optimal Model
"Prompt Augmentation" outperforms pure text entry by reducing cognitive load.
Why Menus Survive
In a GUI, features are
visible affordances. You know you can "Save" because you see a
button.
In an Intent UI (AI), features are invisible. This
creates the
Affordance Crisis. If the user doesn't know what to ask
for, the system's power is inaccessible.
The "Gulf of Envisioning"
Unlike the old "Gulf of Execution" (how do I do
it?), Generative UI creates a new barrier:
"I don't know what this system is capable of
understanding."
The "Blank Page" Paralysis: Users often stare at the blinking cursor, overwhelmed by infinite possibility but lacking specific vocabulary.
5 Core Shifts in Interaction Design
Selected insights from NN/g, Ink & Switch, Microsoft Research, and Harvard Business School.
1. Malleable Software (The "Liquid" UI)
Today, software is rigid. If you want a chart, but the app only has a list view, you are stuck. Malleable Software proposes that the UI should be generated at runtime based on user intent. If you ask for a "Sales Dashboard," the AI shouldn't just retrieve data—it should write the code to render the dashboard layout instantly. The interface is the response.
components
2. The "Sandwich" Workflow
Users reject full automation ("Set it and forget
it") for high-stakes tasks. They prefer a
"Sandwich":
1. Top Slice (Human): Sets
context/strategy.
2. Meat (AI): Generates the
draft/execution.
3. Bottom Slice (Human): Verifies
and refines.
UI must support this verification loop, not hide it.
3. Inverse Information Architecture
Standard IA is "Go to Folder → Find File."
Inverse IA is "I have the File, show me the
Context."
AI flips the navigation model. You start with the
destination (the specific insight or document found
via Intent) and the system must dynamically
reconstruct the "Breadcrumbs" to show you where you
are.
4. The Latency/Fidelity Trade-off
Users perceive "fast and dumb" (Standard GUI) as
reliable, and "slow and smart" (AI) as a utility.
The Threshold: If an AI interaction
takes >1s, it must provide >10x the value of a
click. For simple navigation, 0.1s latency is
mandatory. AI cannot replace navigation until it is
instant.
5. Suggestibility & "Empty States"
The biggest barrier to Intent UI is "What can I
say?" The blank text box causes anxiety.
The solution is
Contextual Chips—dynamic, situation-aware buttons that suggest
intents based on what is currently on screen. It
bridges the gap between Recognition (GUI) and Recall
(AI).
The Articulation Barrier
Experience the friction of "knowing what to say" vs. the effort of "finding where to click."
Intent-Based (AI)
Generative UI / LLM
Command-Based (GUI)
Hierarchical Navigation
The "2-Second Rule"
If articulating a prompt takes longer than clicking a button (approx 2s), users revert to GUI. AI is reserved for "Ambitious Tasks" where manual effort is >10s.
2025 Performance Metrics
Data synthesized from Microsoft New Future of Work, Baymard Institute & NN/g.
Latency Tolerance Thresholds
Users wait longer for AI, but drop-off is steep.
The "Ambitious Task" Paradox
AI excels at complex tasks, fails at trivial ones.
Task Completion Time
Includes cognitive & system latency.
The "Processing" Tax: While input is fast, AI processing latency (3-5s) erodes speed advantages for simple tasks.
Preference by Task Complexity
Users revert to GUI for simple tasks.
The 2026 Implementation Matrix
"Go to Settings."
AI is overkill. Latency
(2s) is unacceptable for a 200ms action.
"How do I change privacy?"
AI suggests the
path, or provides a deep link.
"Summarize Q3 reports and email them."
Impossible
with GUI alone. AI is 10x faster.
Smart Apps
Leveraging Edge AI and FunctionGemma for Zero-Latency, Privacy-First Integration.
The Problem
Clients have massive, complex applications suffering from "Feature Creep." Users are overwhelmed by menus and navigation trees.
The Solution
We do not need to rebuild apps from scratch.
We add an intelligent "Smart Layer" on top of existing codebases.
The Result
- ✓ The app acts as an agent.
- ✓ Complex workflows become simple conversation streams.
- ✓ Preserve existing investment while adding futuristic capabilities.
The Engine: FunctionGemma
270M Parameter Model
Specialized AI: Optimized for action and structure, not just chat.
Deployment Flexibility
- On-Device: Zero latency, runs on phone/laptop.
- Client Server: Hosted on private infrastructure.
Technical Specs
- Footprint: ~300MB model file
- Memory: ~500MB RAM target for on-device inference
- Accuracy: ~85% task success on evaluation suite
The Architecture
Big Players Standards: Google A2UI, Vercel json-render, Thesys — built for Cloud LLMs.
Our Solution: Lightweight protocol tailored for Small Language Models (SLMs) and function calling.
Safety: AI strictly fills forms for approved components. Never writes raw code.
Data Sovereignty
GDPR Ready. Data never needs to leave the browser or private network.
Reliability
Offline Capable. Works in subways, airplanes. Zero Latency (<200ms).
Cost Advantage
Lowers cloud costs by offloading compute to the edge.
The "Smart Layer" in Action
Scenario A: Smart Navigation
AI identifies intent → Deep links directly to BillingForm view.
Saves 5+ clicksScenario B: Generative UI
AI sends component type with props → App renders what is needed.
Instant ReportingScenario C: App Control
AI parses details → Calls Calendar API → Shows confirmation card.
Task in secondsReal-World UI Examples (Concept)
Resources & Demos
Hybrid Interfaces Win
Routine Tasks
"Remind me to call John."
- 3x Faster Execution
- Zero Interaction Friction
Complex Discovery
"Show sales..." [AI Suggests Q1]
- Low Articulation Barrier
- Verifiable Output
Safety Critical
Payments, Deletion, Settings
- 100% Predictable
- Zero Hallucination Risk