From Static to Dynamic
Traditional software interfaces have mostly been made up of clear visual elements like text, forms, drop-downs, and buttons. Think of these as "nouns" on the screen.
With AI, we're moving towards interfaces focused on "verbs" like workflows, auto-complete, auto-suggest, and gathering information. We don't have the tools yet to show these verbs well on screen, which is why there's so much new stuff happening with AI interfaces.
Voice Interfaces
Voice interfaces like Vapy and Retell AI show how important it is to have low latency for smooth interactions. If there's too much delay between when a person speaks and when the AI responds, the conversation feels robotic instead of natural.
These interfaces work best when:
They give visual feedback during voice recognition and response
They keep response times very short (measured in milliseconds)
They handle interruptions smoothly
They adjust to unexpected information (like when "Steve" answered instead of "Aaron" in the debt collection call)
For developers creating these interfaces, sharing technical details like response times helps understand what feels natural versus robotic.
AI Agents and Visual Workflows
As AI agents carry out tasks on their own, we need interfaces to watch and manage them. Gum Loop shows this with canvas-based workflows that display each step an agent will take.
Display the sequence of actions in a visual flow
Use colors to differentiate input types, actions, and outputs
Allow modifications at each step
Work best with complex, branching decision trees
This method is based on traditional flowchart principles but makes them interactive instead.
Grid Interfaces for Data Collection
Answer Grid shows how AI can fill structured data into formats like spreadsheets. Each cell acts as an agent, gathering specific information.
Turning open-ended prompts into structured data
Giving example prompts to help users understand what it can do
Allowing gradual additions of new data needs
Including sources for each piece of information to build trust
Using inline references like those in academic papers
Prompt-to-Output Interfaces
Polymet demonstrates how interfaces can turn descriptive prompts into complex designs. These interfaces are most effective when they:
Offer example prompts to help users get started
Give engaging feedback during generation to make waiting easier
Allow specific parts to be revised without redoing everything
Enable step-by-step improvement of certain elements
The challenge is to balance creative freedom with reliable results, especially since users might not know if their terms (like "glass morphic") will be understood.
Adaptive Interfaces
Zuni shows how interfaces can change based on the content. Instead of displaying all possible options like traditional software with many buttons, adaptive interfaces:
Show only options that are relevant to the current context
Keep consistency with keyboard shortcuts even when visual elements change
Clearly show when keyboard input is for controlling the interface versus typing text
Balance between automation and user control
Video Generation Interfaces
Argil demonstrates how interfaces for AI video creation balance quality and speed.
Offering fast, low-quality previews to refine content
Being clear about the time needed for high-quality results
Allowing detailed control over expressions and camera angles
Combining script editing with visual controls
The Shift to Verb-Based Interfaces
We're seeing a major change in software interfaces. Just like touchscreens changed design in the 2010s, AI is making us rethink how people and machines work together. The focus is moving from fixed elements to active processes → from things to actions.
The best interfaces get people involved while letting AI tackle the tough stuff, giving you just the right controls without overwhelming you with techy stuff. These new trends are popping up quickly, showing just how fast interface design can evolve with new tech.
