What Is Agentic Design?
On this page
Agentic design has two distinct meanings. In AI engineering, it refers to agentic design patterns: architectural patterns like planning, reflection, and tool use for building AI agents. In design and frontend work, it refers to agentic design systems: design systems structured so AI agents can read and apply them, typically via files like DESIGN.md.
The short answer#
"Agentic design" collides two unrelated fields that happen to share an adjective. Which one you mean determines which resources will help you:
| Agentic design patterns | Agentic design systems | |
|---|---|---|
| Domain | AI / agent engineering | Design systems, frontend, brand |
| The question it answers | "How do I architect an AI agent?" | "How do I make my design system usable by AI agents?" |
| Typical searcher | ML and software engineers building agents | Design system maintainers, frontend developers, design engineers |
| Canonical resources | arXiv surveys, Microsoft's agent courses, the Springer Agentic Design Patterns book | This site's agentic design systems guide, the DESIGN.md spec |
This page defines both fairly and points you to the right depth for each. We work in the second field; for the first, we link out to the sources practitioners actually cite.
The collision is recent and understandable. AI engineering settled on "agentic" to describe goal-pursuing systems, and "design patterns" was the natural borrowing from software architecture. Meanwhile, the design systems community needed a word for "a design system agents can consume," and "agentic design system" was the obvious construction, reinforced when Google shipped DESIGN.md in 2026 and gave the idea a concrete artifact. Two communities, one adjective, zero shared subject matter.
Meaning 1: What are agentic design patterns?#
In AI engineering, "agentic" describes systems that don't just answer a prompt but pursue a goal: they plan, act through tools, observe results, and iterate with limited human supervision. Agentic design patterns are the recurring architectural building blocks for such systems, the field's equivalent of software design patterns.
Four pattern families come up in nearly every treatment of the topic:
- Reflection: the agent critiques and revises its own output before finalizing it, e.g. generating code, then reviewing that code for bugs and rewriting it.
- Tool use: the agent calls external tools and APIs (search, code execution, databases) instead of relying only on what the model can produce as text.
- Planning: the agent decomposes a goal into steps, sequences them, and adjusts the plan as results come in.
- Multi-agent collaboration: several specialized agents (e.g. a planner, a coder, a reviewer) coordinate on a task that would overwhelm a single prompt.
Beyond these, the literature catalogs patterns for memory, guardrails, routing, and human-in-the-loop checkpoints. If this is what you came for, the established sources are better reading than we are: the survey literature on arXiv, Microsoft's AI Agents for Beginners course, IBM's explainers on agentic architectures, and Antonio Gulli's Agentic Design Patterns (Springer). Those texts define the vocabulary the AI engineering community actually uses.
Note what this meaning is not about: it says nothing about visual design, brand, typography, or UI consistency. "Design" here means system architecture.
Meaning 2: What is an agentic design system?#
The second meaning lives in the design and frontend world, and it is the one this site covers:
An agentic design system is a design system structured so that AI agents can read, interpret, and apply it autonomously: design tokens, component specs, and usage rules exposed as machine-readable context (via files like DESIGN.md or an MCP server) rather than as documentation written only for humans.
The problem it addresses is concrete. AI agents increasingly write UI code, but a traditional design system (a documentation site, a Figma library, a PDF brand book) is invisible to them. The agent can't open Figma, so it guesses, and the result is off-brand UI with fabricated hex values. An agentic design system closes that gap by publishing the design system in a form agents can consume directly.
The best-known artifact of this approach is DESIGN.md. DESIGN.md is an open-source file format from Google that stores a complete design system in a single Markdown file: machine-readable design tokens (colors, typography, spacing, components) in YAML frontmatter, plus human-readable rationale in Markdown prose. AI coding agents like Claude Code, Cursor, and Copilot read it to generate UI that consistently matches your brand. It is, in one sentence, the file that makes a design system agentic. See what DESIGN.md is for the format itself, and the full agentic design systems guide for the architecture around it (MCP servers included). Core terms are collected in our glossary.
Related terms people confuse#
A third, adjacent field completes the confusion matrix:
- Agentic UX / AX design: designing the user experience of agent-powered products, how a human delegates to an agent, supervises it, builds trust in it, and recovers when it fails. This is interaction design work, distinct from both meanings above. Coverage has been growing in design publications such as Smashing Magazine.
- AI design / generative UI: the broad umbrella for AI producing visual output. Both meanings of agentic design fit under it, which is why the umbrella term is rarely useful in a search query.
- Design tokens: not an "agentic" term at all, but the substrate. The W3C-adjacent token format work at designtokens.org is what agentic design systems build on when they expose values to machines.
Which one are you looking for?#
- You're building an AI agent (architecture, orchestration, memory, tools) → you want agentic design patterns. Start with Microsoft's AI Agents for Beginners course or the Springer Agentic Design Patterns book. Nothing on this site will help you ship a planner.
- You maintain a design system and want AI-generated UI to follow it → you want agentic design systems. Start with our definitive guide to agentic design systems, then the DESIGN.md explainer.
- You're designing the UX of an agent product (chat surfaces, approval flows, trust) → you want agentic UX / AX design, best covered by general design publications.
- You're not sure → ask what your deliverable is. Code that is an agent: Meaning 1. Files that an agent reads to stay on-brand: Meaning 2. Screens a human uses to steer an agent: AX.
Next step: if you landed here from the design side, the depth lives in our pillar guide to agentic design systems: definition, anatomy, implementation options, and a five-step migration path.
Frequently asked questions
What does agentic mean in AI?
The term 'agentic' describes AI systems that pursue goals with limited supervision rather than answering a single prompt: they plan multi-step work, call external tools, observe intermediate results, and revise their approach. The word marks the difference between a model that responds and a system that acts on a task until it is done.
What is an agentic design pattern?
An agentic design pattern is a reusable architectural approach for building AI agents: the field's counterpart to software design patterns. The commonly cited families are reflection (self-critique), tool use, planning, and multi-agent collaboration. They are documented in arXiv surveys, Microsoft's agent courses, and Springer's Agentic Design Patterns book by Antonio Gulli, and concern system architecture, not visual design.
What is an agentic design system?
An agentic design system is a design system structured so that AI agents can read, interpret, and apply it autonomously: design tokens, component specs, and usage rules exposed as machine-readable context (via files like DESIGN.md or an MCP server) rather than as documentation written only for humans.
Is DESIGN.md an agentic design pattern?
No. DESIGN.md has nothing to do with agentic design patterns like planning or reflection. It is a file format: Google's open-source way to store a design system as machine-readable Markdown, and therefore an implementation of an agentic design system. The shared word agentic is the only connection between the two concepts.
What is AX design?
AX (agent experience) design, sometimes called agentic UX, is the practice of designing products where AI agents act on a user's behalf: delegation flows, supervision and approval surfaces, trust and recovery patterns. It is a third field, distinct from agent architecture patterns and from machine-readable design systems, though it often ships alongside both.