DEVOURED
Agentic Design System - From Chatbot to Orchestration
Romina Kavcic argues that design systems are evolving into "agentic" operational infrastructure by 2026, requiring components to become contracts with semantic metadata for AI interpretation and orchestration.
What: Romina Kavcic's article, dated May 10, 2026, posits that design systems are transforming from passive libraries into agent-readable infrastructure. By 2026, Gartner predicts 40% of enterprise apps will use task-specific AI agents. This necessitates treating components as "contracts" embedding intent, constraints, and accessibility, moving beyond AI as a chatbot to an orchestration layer. Figma's 2025 Model Context Protocol (MCP) and existing quality automation tools are foundational, enabling agents to understand design context and automate repetitive tasks.
Why it matters: This article provides a critical, forward-looking perspective on the strategic shift required for design systems in the age of AI, emphasizing structured data and governance over mere generative capabilities, and highlighting the transition from design-to-code tools to general-purpose agents.
Takeaway: To prepare for agentic design systems, start by defining a single component as a detailed "contract" with its intent, rules, and anti-patterns, and add similar intent-based metadata to your five most-used design tokens.
Deep dive
- Design systems are evolving from passive component libraries to operational infrastructure that AI agents can read, interpret, and act upon, with human oversight.
- The shift is from AI as a chatbot to AI as an orchestration layer, coordinating work across tools and workflows.
- An "agentic design system" provides enough structure for AI agents to understand what, why, when, and how components and rules should be applied.
- Existing building blocks include Figma's 2025 Model Context Protocol (MCP) for AI access to design context, automated quality checks (accessibility, visual regression), and internal background coding agents (like Spotify's).
- Future agents are expected to operate with "governed autonomy," proposing changes that humans review and systems validate, reducing invisible drift between design and implementation.
- Critical future developments include design tokens carrying rich metadata beyond just values (e.g., intent, usage rules, accessibility requirements) and documentation becoming executable context for machines.
- Runtime adaptation of UI, though exciting, necessitates strong governance, ethics, and audit trails to prevent UX manipulation.
- Potential pitfalls include agents optimizing locally at the expense of the overall product experience, amplifying existing system weaknesses, and generating confidently incorrect information.
- Figma is seen as evolving into a "design system control surface" for managing components, variables, and reviewing agent-proposed changes.
- Designers will shift from creating variations to defining intent, behavior, constraints, and governance rules.
- The article stresses that "structure beats prompts," advocating for deeply structured components and tokens with semantic meaning to build compounding advantage for AI leverage.
Decoder
- Agentic system: A system where AI agents can autonomously perform tasks, make decisions, and coordinate with other agents or systems, typically with human oversight.
- Model Context Protocol (MCP): A standard or method, as introduced by Figma in 2025, that allows AI tools to directly access structured design context (components, variables, styles) from design files, enabling more informed AI operations.
- Orchestration layer (in AI): A system that coordinates and manages the interactions and workflows of multiple AI agents and other tools to achieve a larger goal, moving beyond simple question-answering.
- Semantic tokens: Design tokens that define purpose or intent (e.g.,
color.background.primary) rather than just raw values (e.g.,color.blue.500), providing more context for AI and developers.
Original article
Full article content is not available for inline reading.
Explore all 40 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi