The Missing Token Layer: Designing for Agentic States
What the mobile shift taught us about designing for a new paradigm — and what design systems need now.
As a UI/UX Designer, I began in 2009 developing mobile apps before ‘Design Systems’ was a term, and App Designer sufficed. Figma didn’t exist and Apple’s storyboard and Apple’s Human Interface Guidelines were my source of truth.
Mobile was an exciting paradigm shift, a handheld phone, camera/video camera, walkman, mini-tv, compass, map, flashlight, the everything hand held device and an developer platform where innovation was directly rewarded through the App Store. Enterprises had to rethink their offerings and presence for this new interface of interaction be it with an app or mobile friendly website.
The past two years have presented a similar paradigm shift with AI and the evolving Agentic Layer. In terms of Design Systems, as a UI/UX designer I consider the integration opportunities for autonomy states within existing interfaces, and opportunities for dynamic UI and what that may look like for a given set of goals, problems and openings ripe for innovation.
What this shift means for Design Systems is still being written — and the most pressing gap I see is the absence of a token language for agentic states. Here are some suggestions:
Agentic Interfaces:
-Intent Entry Container/ Agent Input Surface
Autonomy States:
-Agent Thinking
-Agent Acting
-Agent awaiting confirmation
-Agent partial - requires more input
-Agent interrupted
Agent Output Subcategories:
-small, text, diagram, app, report, action for approval, etc.
Agent Phase History:
-modifiable
-fork point - choose a different path
-with source / confidence
Under Agent History, "fork point" in particular is a concept in an agentic session where the user could have chosen a different path — and crucially, can still go back and choose it. That's a new interaction primitive, not just a token property.
Enterprises that didn't build for mobile correctly in 2010-2012 spent years catching up. The same window is open now, yet it's narrower because of the speed of progression of AI and emerging frameworks.
Brad Frost's Atomic Design gave us a generative grammar for interfaces — atoms, molecules, organisms, templates, pages — a periodic table of UI. It was the right metaphor for its moment: stable elements, predictable bonds, deterministic composition. But the agentic layer doesn't behave like a periodic table. It behaves more like Schrödinger's electron cloud model. An electron doesn't have a fixed position; it exists as a probability distribution across possible states, collapsing into a specific location only at the moment of observation — of interaction. Agentic components are the same: an agent isn't thinking or acting or awaiting confirmation in some resolved, static sense. It occupies a superposition of potential states until the interface — and the user — forces a collapse. The electron cloud model suggests that the next evolution of design systems won't be a new set of components, but a new ontology: one that encodes probability, phase, and autonomy as first-class design primitives. Atoms were the right metaphor for building. Orbitals are the right metaphor for agency.
If your team is building in this space, I'd like to hear what you're seeing.
References:
The Future of Enterprise Design Systems
https://www.supernova.io/blog/the-future-of-enterprise-design-systems-2026-trends-and-tools-for-success?utm_medium=referral&utm_source=claude.ai
Atomic Design by Brad Frost
https://atomicdesign.bradfrost.com/
IBM’s Carbon Design System: Carbon for AI https://carbondesignsystem.com/guidelines/carbon-for-ai/
**NEW**
A comprehensive library of interaction patterns for designing human-centered AI systems.
https://ai-interaction.com/atlas