When an AI project stops organizing itself around “models” or “tools” and instead structures around “departments” and “roles,” it signals a landmark shift: AI Agents are entering enterprise-grade applications.
The Agency (github.com/anthropics/agency-agents) surpassed 50K GitHub stars within two weeks of launch. It’s not a prompt template collection or a single Agent wrapper—it’s a complete organizational structure with 147 specialized AI Agents spanning 12 business departments.
The Problem It Solves
Current AI Agent usage is still “single-point”: Coding Agents for code, content generation Agents for marketing, conversational Agents for customer support. But real companies don’t work that way—departments have collaboration workflows, information flow, and quality gates.
The Agency’s approach: organize AI Agents like a company.
12 Departments, 147 Professional Roles
The project covers:
- Engineering - Software development, code review, architecture design
- Design - UI/UX design, visual solutions
- Marketing - Content marketing, SEO, social media
- Product - Product planning, requirements analysis
- QA - Testing strategy, automated testing
- Support - Customer support, knowledge base maintenance
- Spatial Computing - Spatial computing tasks
- And additional business departments
Each Agent has clearly defined responsibilities, input/output specifications, and collaboration interfaces. The Chinese version (agency-agents-zh) extends to 211 plug-and-play AI expert roles across 18 departments, including 46 China-market original agents (adapted for Xiaohongshu, Douyin, and other local platforms).
Quick Start
# Install to Claude Code
# Add agency-agents configuration to .claude directory
# Chinese version
git clone https://github.com/agency-agents/agency-agents-zh.git
cd agency-agents-zh
# Follow README for one-click installation to Claude Code/Cursor/CopilotPrimary Sources: