When AI Learns to 'Draw' Interfaces: How A2UI is Redefining Human-Computer Interaction

Have you ever wondered why, when we talk to AI, it can only "say" things but can't actually "do" anything?

This seemingly simple question reveals a fundamental limitation of current AI interaction—no matter how intelligent AI becomes, its means of expression remains stuck at the "text" level. The emergence of A2UI is breaking through this limitation.

Starting with an Everyday Scenario

Suppose you tell an AI: "Help me book a flight to Shanghai for tomorrow."

How does today's AI respond?

It gives you a wall of text: flight information, price comparisons, things to note... then tells you "please visit such-and-such website to complete your booking."

You need to:

• Read through all that text
• Copy the flight number
• Open another website
• Manually fill in your information
• Complete the payment

AI said a lot, but it didn't actually "do" anything.

This "talk but don't act" pattern exists in almost every interaction we have with AI. Whether you're asking AI to analyze data, fill out forms, or generate reports, you ultimately need to leave the conversation interface and complete the actual operation somewhere else.

AI is like a knowledgeable consultant who never lifts a finger—it can tell you how to do something, but it can't do it for you.

What If AI Could Directly 'Draw' the Interface?

Imagine a different scenario:

You say: "Help me book a flight to Shanghai for tomorrow."

AI directly displays a real booking interface right in the conversation:

• Three flight options in a clear card layout, each card showing airline, departure time, arrival time, and price
• Prices and times at a glance, with options to sort by price or time
• Click the "Select" button, and the card highlights
• A passenger information form appears where you can fill in details directly
• After confirmation, a payment interface appears
• Done

The entire process happens without leaving the conversation interface, without copying and pasting, without switching between multiple apps.

From "saying" to "doing" — this is what A2UI aims to achieve.

This isn't a scene from a sci-fi movie—it's the interaction paradigm defined by the A2UI protocol that Google released at the end of 2025. This technology is moving from concept to reality, and we've already implemented its first complete version.

What is A2UI?

A2UI, short for Agent-to-User Interface, is an open protocol officially released by Google in December 2025.

Simply put, it defines a set of standard rules that allow AI to output interactive interface elements, not just text.

Traditional AI Output vs A2UI Output

Interface Elements Supported by A2UI

These interface elements cover most application scenarios:

Basic Components

• Text: supports rich text formatting, heading levels, emphasis styles
• Buttons: configurable styles, icons, click actions
• Input fields: single-line, multi-line, password, search, and more
• Checkboxes and radio buttons: support grouping and linking

Layout Components

• Cards: containers for aggregating information, with header, content, and action areas
• Lists: vertical or horizontal arrangement, with pagination and lazy loading
• Tables: with sorting, filtering, and pagination
• Tabs: multi-view switching

Data Visualization

• Charts: line, bar, pie, scatter plots, and more
• Progress bars: linear or circular
• Stat cards: key metric displays

Media Components

• Images: with carousel, zoom, lazy loading
• Video and audio: embedded players
• Files: upload and download

Advanced Interactions

• Modal dialogs: confirmation, forms, detail displays
• Dropdown menus: selection and filtering
• Date pickers: single date or range selection

AI is no longer just a "text generator" — it becomes an "interface designer."

What Does This Mean for Different Roles?

For Regular Users: Goodbye to the "Copy-Paste" Era

Previously, talking to AI was like communicating with an assistant who could only talk but never act. You ask it questions, it gives you answers, but the answers are just text. You need to execute the operations described in that text yourself.

It's like asking someone "how do I get to the train station" — they give you detailed directions, but you still have to walk there yourself.

With A2UI, AI can directly help you:

Scenario 1: Filling Out Complex Forms

Before: AI tells you what information to fill in, you find the form, fill it out item by item.

Now: AI directly generates the form, you fill it out in the conversation, AI submits it for you.

Scenario 2: Data Queries

Before: AI tells you what the data is, you need to organize it yourself or export to Excel.

Now: AI directly generates charts and tables, you can click to filter, sort, and export.

Scenario 3: Booking Services

Before: AI tells you available time slots, you go to a calendar app to create an event.

Now: AI displays a calendar component directly, you click to select, booking complete.

Less copying and pasting, less switching between apps — everything happens within the conversation.

For Businesses: A Qualitative Leap in AI Customer Service

Traditional AI customer service has a fundamental problem: it can only answer questions, not solve problems.

When a user says "I want to return this item," traditional AI replies: "Please log into your account, go to the orders page, click the return button..." Then the user has to go do it themselves. If the user encounters any problems, they have to come back and ask again.

This isn't real service — it's just a talking instruction manual.

With A2UI, AI customer service can:

Directly Display Order Lists

User says "I want to return something," AI directly shows the user's recent orders in card format, each order showing product image, name, and purchase date. User clicks the one they want to return — no need to remember order numbers.

Generate Return Application Forms

After user selects an order, AI directly shows return reason options, refund method choices, and a supplementary notes input field. User fills it out and clicks submit.

Display Processing Progress in Real-Time

After submission, AI shows a progress bar: "Application Submitted → Merchant Review → Awaiting Return Shipment → Processing Refund." User can come back anytime to check progress.

From "Q&A bot" to "service terminal" — users don't need to leave the conversation interface to complete the entire business process.

For Developers: Lowering the Barrier to AI Application Development

Previously, to get AI to output structured content, developers needed to:

1. Design complex prompts to get AI to output specific JSON formats
2. Write parsing code to handle various exceptions in AI output
3. Design frontend components to render this data
4. Handle user interactions and feed results back to AI
5. Debug constantly because AI output formats are often unstable

The entire process was full of uncertainty, with long development cycles and high maintenance costs.

With A2UI, developers can:

Use Standardized Protocols

A2UI defines clear data structures — what format AI outputs, what fields it includes, how to describe interactions — all clearly specified. No need to design and maintain proprietary formats.

Use Ready-Made Renderers

Our open-source @zhama/a2ui provides a complete React renderer. Just pass AI output to the component, and interfaces render automatically with interactions handled.

Focus on Business Logic

Developers no longer need to worry about "how to get AI to output the right format," "how to render this data," or "how to handle user clicks." These are all solved by the protocol and renderer. Developers only need to focus on: what the user wants to do and how AI should help.

Lower development barriers, faster AI application deployment. Work that previously took weeks might now take just days.

Real-World Applications

Intelligent Training Systems: Everyone Gets a Personal AI Tutor

The problem with traditional online courses is the "one-size-fits-all" approach — everyone watches the same videos, does the same exercises, follows the same sequence.

But everyone has different foundations, learning speeds, and interests. Someone with programming experience and someone with zero background clearly shouldn't learn the same content.

AI training systems powered by A2UI can achieve:

Dynamic Course Generation

AI first asks a few questions to understand your background and goals, then generates a personalized learning path. Not pulling from a question bank, but generating content in real-time based on your situation.

Interactive Case Studies

When learning business cases, AI doesn't just show text descriptions — it generates an interactive scenario:

• Displays the company's key metrics (charts)
• Provides several decision options (buttons)
• You click an option, AI shows the consequences of that decision
• You can backtrack, try different choices, understand the impact of each decision

This is much more profound than simply reading case descriptions.

Real-Time Progress Tracking

AI generates your skill map, using radar charts or bar charts to show your mastery level in each knowledge area. You can click on a weak area, and AI immediately generates targeted exercises.

Intelligent Data Analysis: Conversational Business Intelligence

The problems with traditional reports are "lag" and "rigidity" — a fixed-format report generated weekly or monthly that might be outdated by the time you see it. And if you want to analyze from a different angle, you need to request IT department help.

AI analysts powered by A2UI are completely different:

Natural Language Queries

You say "Analyze last month's sales in the East region, grouped by product category," AI directly understands your request, queries the data, and generates results.

Dynamic Visualization

AI doesn't return a bunch of numbers — it directly generates the most appropriate charts for the data:

These aren't static images — they're interactive charts generated by AI in real-time based on data. You can:

• Hover to see specific values
• Click on a data point to drill down into details
• Switch time ranges or filter conditions
• Export data or charts

Discover Insights and Take Action

When AI discovers anomalies or trends in the data, it doesn't just tell you "March sales dropped 15%" — instead it:

1. Highlights this anomaly on the chart
2. Generates a list of possible causes (clickable to expand details)
3. Provides several suggested action plans (in button form)
4. You click on a plan, AI further shows execution steps or generates related task forms

Data analysis becomes a conversation, not a static report waiting to be read.

Intelligent Customer Service & Sales: Bots That Can Actually "Get Things Done"

What's the most frustrating experience with traditional chatbots? "Your question has been recorded. Transferring to human agent."

The subtext of this message is: I understood what you said, but I can't help you.

AI customer service powered by A2UI can actually solve problems:

Product Inquiries

User asks "What water purifiers are good for home use?" AI doesn't just list a few product names — instead:

• Generates a product comparison table with key parameters like price, filter life, and suitable water quality
• Table supports sorting (by price, by sales)
• Each row has a "View Details" button that expands to show detailed product information
• Has "Add to Compare" checkboxes to select products for a dedicated comparison card

After-Sales Service

User says "My order hasn't arrived yet," AI:

• First displays the user's pending delivery orders (automatically retrieved from the system)
• User clicks on the problematic order
• AI shows logistics tracking timeline (each node clickable for details)
• If there's indeed a problem, generates operation buttons like "Rush Delivery" or "Apply for After-Sales"
• User clicks and the operation completes directly, no redirect needed

Booking Services

User says "I want to schedule weekend installation," AI:

• Displays calendar component with available time slots highlighted
• User clicks to select a time
• Generates confirmation card: time, address, contact info, notes
• User confirms and booking is complete, reminder automatically sent

Customer service bots can finally "get things done" — not transferring to humans, not making users do it themselves, but directly completing tasks for users.

Why Now?

A2UI didn't appear out of nowhere — it's the natural product of multiple technologies maturing.

Large Language Model Breakthroughs

Large language models like GPT-4, Claude 3, and Gemini can now:

• Understand complex, multi-step instructions
• Generate well-formatted structured output (JSON, XML, etc.)
• Adjust output content and format based on context
• Handle multi-turn conversations, maintaining state and memory

This means having AI output interface descriptions conforming to the A2UI protocol is entirely feasible. AI doesn't need to "learn to draw interfaces" — it just needs to output data describing interfaces, and the renderer handles the "drawing."

Frontend Technology Maturity

The core idea of modern frontend frameworks like React, Vue, and Svelte is "data-driven interfaces" — you give the framework data, it renders it into an interface.

This perfectly aligns with A2UI's philosophy. AI outputs data, frontend frameworks render interfaces. All that's needed between them is a "translation layer" — this is the @zhama/a2ui renderer we developed.

Rising User Expectations

After years of AI adoption, users have become accustomed to talking with AI. But they're increasingly unsatisfied with "just talking."

• Users want AI to help them complete tasks, not just provide information
• Users want more direct operations, not jumping between multiple apps
• Users want more natural interaction, like communicating with a person

Market demand exists, and technical conditions are mature.

A2UI appeared at exactly this time to connect AI capabilities, frontend technology, and user needs.

What Are We Doing?

When Google released the A2UI protocol, they announced the official React renderer would be released in Q1 2026.

But we couldn't wait.

After the protocol was released, the Zhama AI team immediately began implementation work. Based on the A2UI protocol specification, we developed the first complete React implementation: @zhama/a2ui.

Why Implement Early?

Technical Validation

We believe A2UI is the right direction, but whether the protocol can be implemented, whether the experience is smooth, whether there are hidden issues — you only know by actually building it. Through implementation, we validated the protocol's feasibility and discovered and fed back areas needing improvement.

Accumulating Experience

Waiting until the official renderer releases to start learning means you're already behind. By implementing early, our team gained deep understanding of A2UI's design philosophy and technical details, accumulating valuable experience for future application development.

Serving Customers

Our customers have real needs — they want to use this technology in their products as soon as possible. Waiting isn't an option; we chose to take initiative.

Features of @zhama/a2ui

Complete Implementation

100% coverage of the A2UI protocol specification, including all component types, data binding, event handling, and action systems. Not a demo — a production-ready complete implementation.

Independent Styling System

Comes with complete CSS styles, no dependency on Tailwind or other frameworks. Uses a2- prefix to avoid style conflicts with host applications. Supports automatic dark mode.

Security-First Design

AI outputs declarative data, not executable code. The renderer only renders predefined components, won't execute arbitrary code — no code injection risk.

Open Source and Free

Open source under MIT license — you can freely use, modify, and distribute. We believe open source leads to better technology development.

We believe that embracing new technology early means gaining competitive advantage early.

What Does the Future Hold?

A2UI is just the beginning — it opens up an entirely new interaction paradigm. As technology develops, we might see more exciting changes.

Multimodal Fusion

Currently A2UI mainly handles the "AI generates interface → user clicks to operate" flow. In the future, input methods will become richer:

• Voice commands: you speak, AI adjusts the interface
• Gesture operations: in AR/VR environments, interact with AI-generated interfaces using gestures
• Eye tracking: AI adjusts information display based on your gaze

Interfaces will no longer be static but will adapt in real-time to your behavior.

Extreme Personalization

Today's applications show the same interface to everyone for the same function. But everyone has different preferences — some like information-dense layouts, others prefer clean and simple; some are visual types who prefer charts, others prefer text lists.

With AI-generated interfaces, this personalization becomes possible:

• Same "view sales data" function: analysts see detailed multidimensional charts, CEOs see a one-page concise summary
• Interface style automatically adjusts to user preferences
• Information presentation order optimizes based on user history

No-Code Development Goes Mainstream

Currently, developing an application interface requires professional designers and engineers. But if AI can generate interfaces:

• Product managers describe requirements in natural language, AI generates prototypes
• Entrepreneurs without technical backgrounds can quickly validate ideas
• Internal business systems can be customized by business users themselves

Product prototypes shrink from weeks to minutes, dramatically lowering the barrier to software development.

Blurred Human-Machine Boundaries

Ultimately, the boundary between "applications" and "AI conversations" will become blurred. You might not be able to tell whether you're "using an app" or "talking to AI" — because the two merge into one.

Interface is conversation, conversation is interface. Every sentence you say might change the interface; every click you make is a communication with AI.

The Evolution of Interaction Methods

Looking back at the history of computer interaction methods, every revolution has brought massive industry transformation:

Command Line Era (1970s-1980s)

Users interacted with computers by typing text commands. Required memorizing numerous commands, extremely high barrier — only professionals could use them.

Graphical Interface Era (1980s-2000s)

The appearance of mice, windows, and icons allowed ordinary people to use computers. This sparked the explosion of the personal computer industry, with Microsoft and Apple becoming giants.

Mobile Internet Era (2010s)

Touchscreens made interaction more intuitive, app stores made software distribution more convenient. This sparked mobile internet prosperity, with smartphones becoming essential for everyone.

Conversational AI Era (2020s)

Large language models like ChatGPT made human-machine dialogue possible. But conversation output remained text, with limited capabilities.

A2UI Era (2025-)

AI can output interactive interfaces, truly moving from "saying" to "doing." What industry transformation will this spark? We're witnessing it and participating in it.

Final Thoughts

From command line to graphical interfaces, from web pages to mobile apps, every revolution in interaction methods has created enormous commercial value and birthed new tech giants.

A2UI represents the next revolution: from "humans adapting to machines" to "machines adapting to humans."

In this new paradigm:

• AI is no longer just a tool that answers questions — it becomes an intelligent assistant that understands needs, generates interfaces, and completes operations
• Applications are no longer fixed interface flows — they become dynamic interaction experiences generated based on user needs
• Development is no longer the exclusive domain of professional engineers — it becomes a creative activity anyone can participate in

This isn't science fiction. This is happening now. And we're standing at the starting point of this revolution.

Want to learn more or get started? Visit our GitHub repository, or contact us to discuss how to apply this technology in your business.

This article was written by the Zhama AI technical team. We are committed to making AI technology better serve humanity.