Adaptive User Interface

From Static to Adaptive

Picture this: a design system that doesn’t just store buttons and cards—it builds your interface for you. Real-time. Personalized. Imagine logging into your favorite app, and it’s already set up just the way you like it. Big text? Done. Minimal layout? Handled.

So, adaptation means modifying an interface to benefit the user. Currently, we adapt to interfaces.

But how do I imagine this would work? For AI to learn, we need to go through 4 phases:

1 Understand phase

This is where we lay the foundation by gathering and interpreting user data. Think of it as listening carefully to every user action:

• What they click, tap, or scroll.

• When they visit—morning, night, or in between.

• How they interact—adjusting text size, toggling dark mode, or skipping features.

• Where they spend the most time.

• Why they behave the way they do, inferred from patterns.

2 Summarize phase

Analyzing and categorizing data into Foundations, UI Components, Design Systems, ML Models, User Segments, Personalized Rules, Behavioral Patterns, Metrics Analysis, and User Feedback.

• "Oh, they always make the text bigger”

• "They check weather every morning before"

• "They really like looking at pictures more than reading"

• "They check stock market”

3 Predict phase

Putting things in context for Usage Patterns, User Preferences, Layout Preferences, Color, Content Adaption, Interaction Patterns, Typography Needs, Accessibility, Time Context, and Performance.

They probably want to see the weather report first and prefer dark mode at night.

4 Generate phase

Finally, the website becomes generated based on your preferences. Think of this as a smart LEGO set. Instead of hard-coded elements, these building blocks can change positions, stretch, shrink, and reshape themselves.

For the first person (User 1): "This person likes to get the weather info first thing in the morning, so let's give them:

• A minimal header

• Weather

• Stats

• Personalized widgets

• Footer

For the second person (User 2): This person likes to read and explore, so let's give them:

• A big header

• Lots of content for reading

• Their profile

Why I think this is possible soon

A decade ago, AI could barely spot a cat in a photo.
Four years ago, GPT-2 struggled to write coherent sentences.
Today, GPT-4 writes code, solves math, and aces college exams.

Progress feels shocking—until you look at the trendlines.
Deep learning has been on a straight path: scale up, models get smarter. That’s it.

Rough estimates of past and future scaleup of effective compute (both physical compute and algorithmic efficiencies), based on the public estimates discussed in this piece. As we scale models, they consistently get smarter, and by “counting the OOMs” we get a rough sense of what model intelligence we should expect in the (near) future. (This graph shows only the scaleup in base models; “unhobblings” are not pictured.)

Image from Leopold Aschenbrenner

Estimates by Epoch AI of algorithmic efficiencies in language modeling. Their estimates suggest we’ve made ~4 OOMs of efficiency gains in 8 years. Image from Leopold Aschenbrenner.

So, in short:

• AI systems are getting much smarter and cheaper to run

• What used to cost $1,000 now costs $1

• AI can now handle complex tasks like high-school-level math

• These improvements are happening consistently and rapidly

These efficiency leaps in AI models make adaptive interfaces not just possible but affordable and scalable. 🚀

What This Means

For Users

• No more hunting for features you use often

• Comfortable reading experience without manual adjustments

• Personalized layouts that make sense for you

• Time-saving automated adjustments

For Designers

• Creating flexible systems instead of fixed layouts

• Building components that can adapt

• Setting rules rather than exact specifications

• Testing for various scenarios instead of one fixed design

For Developers

• Dynamic, Rule-Based Development

• Integration of AI and ML Models

• Handling Data Responsibly

• More Modular Code

• More complexity

🔥 Challenges

• The biggest challenge is understanding users. Even as humans, we struggle to connect with them. So, how can we expect AI to bridge that gap?

• Interfaces must avoid overwhelming the user with constant change, as this increases cognitive load and confusion.

• User data can be noisy (e.g., accidental taps) and hard to interpret.

• Needs good-quality data to work well.

• We must think about ethical implications and user privacy.

• Needs robust testing systems.

✅ The Takeaway

The old way: Design systems are libraries of fixed components.

The new way: Design systems are living, breathing organisms that react, change, and improve.

The result? Happier users, faster workflows, and interfaces that feel custom-built just for you.

The models? More material to learn, better outcome.

The future isn’t static. It’s adaptive.

But I’m not sure we are ready for this shift. Yet! 😅

Read more about AI

🔗 From GPT-4 to AGI: Counting the OOMs from Leopold Achenbrenner

---

💎 Community Gems

The state of AI in design systems from Supernova

🔗 Link

The Future of Design Tokens - Online Jam with Tokens Studio
on 🗓️ Wed, Jan 15, 2025, 7:00 PM - 8:30 PM (your local time)

Organizer: Into Design Systems Community & Conference

🔗 Link

Variables to Code Figma plugin

🔗 Link

Figma to Code Figma plugin

Figma to Code (HTML, Tailwind, Flutter, SwiftUI)

🔗 Link

Componly

Track your design system adoption

🔗 Link

---

❤️ People to follow

This week, I would like to introduce you to Jan Toman 🙌

He is the Director of Design at Supernova, bringing a unique, multi-dimensional perspective to product development. With a background as a developer, product manager, and UX designer, he understands the nuances of each stage in the process. His passions lie in design systems, product management, design tooling, and automation — but above all, he’s driven by a mission to connect the disconnected.

📹 Mastering the art of crafting code-aligned UI kits - Jan Toman (Schema 2021)

🔗 Video

📹 Continuous design systems - Jan Toman at Into Design Systems Conference

🔗 Video

🔗 Follow Jan on Linkedin

---

🙋‍♀️🙋🏽‍♂️ Questions from the community

If you have any questions, email me or add them to the comment below. 🙌