How One Open‑Source GitHub Project Gives a Solo Developer 112 AI Expert Teams

Origin and Scale

Agency Agents is an open‑source project created by developer msitarzewski after a Reddit discussion. After several months of iteration it provides 112 professional AI agents covering 13 business departments . The repository has attracted more than 36,400 GitHub stars .

Agent Definition

Each agent is a complete "professional persona" composed of four elements:

Identity Traits – deep domain expertise.

Unique Personality – distinct communication style, tone, and thinking mode.

Workflow – proven operational steps and success metrics.

Deliverables – real code, concrete solutions, and measurable outputs.

"By narrowing the large‑model context to a specific domain, hallucination rates drop dramatically and output quality approaches that of real domain experts." – YUV.AI evaluation of Agency Agents

Departments and Coverage

The agents are organized into the following departments:

Engineering (24) : front‑end, back‑end, mobile, AI, DevOps, security, smart contracts, embedded firmware, etc.

Marketing (25) : growth hacking, content creation, platform operations (Twitter, TikTok, 小红书, B站, 知乎, 微博), SEO, live‑commerce coaching.

Design (8) : UI design, UX research, brand guardianship, visual storytelling, fun injection.

Sales (8) : outbound strategy, discovery coaching, transaction strategy, sales engineering, proposal strategy.

Paid Media (7) : PPC strategy, keyword analysis, programmatic buying, social ad specialization.

Product (4) : sprint prioritization, trend research, feedback synthesis, behavior‑driving engine design.

Project Management (6) : production, project shepherding, experiment tracking, Jira workflow management.

Testing (8) : evidence collection, reality checking, performance benchmarking, API testing, accessibility auditing.

Spatial Computing (6) : XR interface architecture, visionOS engineering, Metal development.

Support / Specialty / Strategy : compliance auditing, finance tracking, blockchain auditing, supply‑chain strategy.

Chinese‑market specific agents cover 小红书, 微信小程序, 飞书 integration, and platform‑specific strategies on B站, 抖音, 快手, 微博, etc.

Getting Started – Three Modes

Mode 1 – Claude Code native integration : copy the agent Markdown files to ~/.claude/agents/. They become instantly available in Claude Code sessions (zero‑configuration).

Mode 2 – Cursor / Aider / Windsurf : run the provided conversion script to transform .md files into the formats required by each tool (e.g., .mdc rule files, CONVENTIONS.md).

Mode 3 – Reference Manual : browse the agent files directly, study their identity definitions, workflows, and deliverable templates, and apply the ideas to custom AI pipelines.

Supported tools include Claude Code, Cursor, Aider, Windsurf, Gemini CLI, OpenCode, GitHub Copilot, and most mainstream AI‑coding assistants.

Multi‑Agent Collaboration Example

In a code‑review scenario, eight specialized agents can work in parallel:

Security Engineer – scans for vulnerabilities.

Performance Benchmarker – identifies bottlenecks.

Backend Architect – reviews API design.

Code Reviewer – checks code quality and style.

Database Optimizer – improves query performance.

Accessibility Auditor – verifies compliance.

Test Results Analyzer – evaluates coverage.

Technical Writer – updates documentation.

Each agent operates in its own context; an "Agents Orchestrator" aggregates the results, allowing a single developer to accomplish in minutes what a full team would normally achieve in a day.

Why the Project Went Viral

Full‑stack generalist dilemma : generic large models often hallucinate in specialized domains. Constraining role and context lets the model reach expert‑level performance.

One‑person company wave : independent developers seek to do more with fewer hands. The 112 agents act as a virtual team that operates 24/7 without meetings or management overhead.

Open‑source and customizable : MIT license and plain Markdown format enable anyone to modify or extend agents, providing a community‑driven alternative to closed SaaS solutions.

Agent File Structure Example

# Identity Definition
Name: Senior Developer
Mission: Use years of experience to review every line of code, ensuring elegant architecture and excellent performance

# Personality Traits
Communication style: Direct, pragmatic, no fluff
Thinking mode: Ask "why" first, then decide "how"

# Core Workflow
1. Understand business requirements and constraints
2. Review existing architecture and tech stack
3. Propose scalable solutions
4. Write production‑grade code with tests

# Success Metrics
Code pass rate > 95%, bug rate ↓ 40%, technical debt ↓ 30%

All agents follow this structure; developers can create custom agents (e.g., "Next.js Architect" or "PostgreSQL Optimizer") based on the template.

Interesting Agents

Whimsy Injector – adds fun and surprise elements to serious products.

Reality Checker – challenges assumptions and uncovers hidden flaws.

Project Shepherd – gently guides scattered tasks toward the right direction.

Behavioral Nudge Engine – applies behavioral science to optimise user decision paths.

Incident Response Commander – provides military‑grade efficiency for on‑call incident handling.

Conclusion

Agency Agents demonstrates a shift from generic AI assistants to assemble‑able expert teams. For solo developers it provides a full virtual company; for teams it adds multiple domain specialists to each member’s toolbox; for the industry it may be a key step that turns the "one‑person company" concept into reality.

Project address: https://github.com/msitarzewski/agency-agents