ai agents⁽²⁵⁾

Natural language replaces menu-driven navigation. Queries like 'Show me what I can afford next month' become first-class product interactions — an intent layer that rewrites how financial and consumer products are designed. Explores autonomy/automation balance and generative UI patterns for agent-first product surfaces.

Open-source React 19 design system from Meta with 150+ accessible components, seven themes, templates, and a CLI. Shared conventions and documentation support people and coding agents.

Collection of copyable interface primitives for AI products, including streaming text, approval cards, tool-call states, task progress, retrieved context, and data diffs.

Emerging design practice: formal 'behavioral contracts' define how AI agents decide, handle failures, and communicate. These policy documents live alongside wireframes and specs — a new design artifact for the agentic era. Marks the beginning of a discipline where designers define agent constraints, not just visual flows.

Design MCP server and shadcn/ui registry that gives coding agents design tokens, UI principles, component code, icon direction, and accessibility review rules.

MCP server that lets AI assistants search curated design system knowledge, including WCAG, ARIA, W3C standards, and major public design systems.

Open-source format from Google Labs that describes a design system to coding agents. Combines YAML design tokens with markdown rationale in one file, plus a CLI to lint, diff, and export tokens to Tailwind or DTCG.

A paradigm shift for product design: as AI agents increasingly initiate transactions, interfaces must be discoverable and executable by machines, not just humans. Explores new design constraints for financial products, e-commerce, and any service where agents act on behalf of users.

Autonomous agents now span Figma → Jira → GitHub — detecting design drift, enforcing brand principles, and flagging accessibility failures in code. This is the governance-beyond-humans model: agents become the system's immune system. Unified collaboration hubs replace handoff lag with real-time spec access in editors and tickets.

Official open-source index of Figma plugins, widgets, agent skills, and developer references. Useful for finding working examples and source code for extending Figma and its MCP workflows.

Open-source bridge between AI agents and Figma that reads selections into framework-aware code and writes changes back to the canvas through a local MCP server and Figma plugin.

Products are shifting from 'user executes tasks' to 'user supervises AI agents.' This requires entirely new UX patterns: confidence levels, decision logic visibility, and human override points. Design moves from task flows to orchestration interfaces — trust and observability become the new UX primitives.

Native macOS canvas for running multiple coding agents in spatial terminal nodes. Connect agents, assign reusable roles, draw diagrams, capture notes, and isolate work in APFS-backed floors.

How design teams scale via role-specific agents (strategist, visual designer, accessibility reviewer) coordinating through shared docs. Agents handle consistency checks; humans preserve judgment on aesthetics. A practical model for augmenting design orgs without losing decision-making authority.

Real-world multi-agent product pipeline: PM agent writes PRD → Design agent prototypes → Eng agent generates scaffold. With human review checkpoints, end-to-end orchestration compresses 6 weeks of work to 20 minutes. Demonstrates practical agent chains for product delivery.

Open-source harness that runs coding agents against a React design system and grades component use, API fidelity, tokens, accessibility, and compilation. Compares results across models and context levels.

Open-source, AI-native vector design canvas with prompt editing, diffable JSON files, MCP access, and code export for React, SwiftUI, Flutter, and other frameworks.

Native macOS bridge that gives MCP-compatible AI assistants access to Calendar, Mail, Notes, Reminders, Messages, Maps, and more. Automation runs locally through Apple apps.

Vector design canvas for desktop apps and IDEs that stores open .pen files in the repo, imports Figma work, and supports bidirectional design and code changes through local MCP tools.

Native macOS menu bar tracker for usage, costs, and rate limits across Codex, Claude, Cursor, Copilot, Gemini, and other coding tools. Useful for planning long agent sessions.

Official shadcn MCP server for browsing, searching, and installing registry components from AI assistants. Works with public, third-party, and private registries.

AI Agents workflows powered by Anthropic, xyflow, and AI SDK with one-click installable examples via shadcn CLI

Source-linked field study of how 20 maintained design systems expose MCP servers, agent skills, llms.txt, editor rules, registries, and other machine-readable interfaces.

Declarative motion graphics for AI agents

Practical framework for building AI agents for design teams: specialized vertical agents (UX audit, feasibility assessment, market intelligence) outperform horizontal generalist chatbots by 10x. Shifts the conversation from 'add ChatGPT' to 'build domain-specific agents.'