Beyond the Glass Slab
The Era of Malleable Software
Current apps are static factories. Future apps are liquid.
Exploring the 5 critical shifts in interaction design driven
by Generative AI.
Deep Dive: 5 Core Shifts
Selected insights from NN/g, Ink & Switch, and Microsoft Research.
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).
Lab: The Articulation Barrier
Comparing the cognitive load of "Chatting" vs. "Clicking" for simple tasks.
Chat-Only
High FrictionContextual (Hybrid)
OptimalAdoption Metrics
Preference by Task Complexity
Users revert to GUI for simple tasks.
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.