# Your AI Dream Team Is Already Assembled

- Author: shahbazsyed (https://wurk.fun/user/shahbazsyed)
- Published: 2026-04-03
- Canonical (HTML): https://wurk.fun/blog/your-ai-dream-team-is-already-assembled
- Cover image: https://ik.imagekit.io/wurk/Gemini_Generated_Image_d0a4prd0a4prd0a4_xyFSfpAaP.png

---

There's a moment every power user of AI assistants knows well you're deep in a coding problem and the same model that just helped you architect your database is now trying to write you a TikTok caption. It works but it doesn't feel *right*. You'd never ask your lead architect to handle community management. So why do we ask our AI to?

That's the insight behind **agency-agents**, a GitHub repository by [@msitarzewski](https://github.com/msitarzewski) that has exploded to over **60,000 stars** since it was born from a Reddit thread. The premise is elegantly simple what if instead of one AI trying to be everything you had a full agency of specialized experts each with their own personality, processes and proven deliverables?

> *"Think of it as assembling your dream team, except they're AI specialists who never sleep, never complain and always deliver."*

---

## What Is agency-agents, Really?

At its core agency-agents is a collection of **meticulously crafted prompt files**. Each file defines an AI "agent" not a separate model or API but a personality and set of instructions that transforms how Claude (or other supported AI tools) behaves.

These aren't generic "act as a developer" templates. They're detailed specifications covering identity, mission, critical rules, deliverables, workflows and success metrics. When you copy these agent files into your Claude Code setup you can activate any of them during a session and the AI shifts into that specialist's mode entirely.

The repository is structured as a proper organization, with agents organized into **12 divisions** that would look right at home on an actual agency's org chart.

---

## A Tour of the Divisions

With **147 agents** spread across 12 divisions, the scope is remarkable. Here's what's inside:

### 💻 Engineering Division — 23 Agents
From Frontend Developer and Backend Architect to a Solidity Smart Contract Engineer and Embedded Firmware Engineer. Includes rare finds like a Threat Detection Engineer, an SRE, a Git Workflow Master and an AI Data Remediation Engineer for fixing broken pipelines at scale.

### 🎨 Design Division — 8 Agents
Includes the expected UI/UX specialists but also the delightful **Whimsy Injector** whose entire mission is adding personality and delight to products, and an Inclusive Visuals Specialist focused on culturally accurate unbiased imagery.

### 📢 Marketing Division — 27 Agents
The largest division. Platform specific experts for TikTok, Instagram, Reddit, LinkedIn and a full suite of Chinese market specialists: Xiaohongshu, Weibo, Bilibili, Douyin, Kuaishou. There's even an **AI Citation Strategist** focused on improving brand visibility inside LLM responses (GEO/AEO).

### 💰 Sales & Paid Media Divisions
Seven focused sales agents (Outbound Strategist, Deal Strategist, Pipeline Analyst) and seven paid media experts covering PPC, programmatic, paid social and tracking. Notably tactical the Deal Strategist knows MEDDPICC qualification cold.

### 🎮 Game Development Division
Covers Unity, Unreal Engine, Godot, Blender and Roblox with engine-specific agents like a Unity Shader Graph Artist, Unreal World Builder and Roblox Experience Designer. Genuinely useful for indie developers.

### 🎯 Specialized Division — 28+ Agents
The wildcard section. An Agents Orchestrator, Blockchain Security Auditor, French Consulting Market Navigator, Korean Business Navigator, Compliance Auditor and even a Study Abroad Advisor. This is where the community's creativity really shows.

---

## The Philosophy: Personality Over Prompts

What separates agency-agents from a basic prompt library is the **depth of agent design**. Each agent isn't just a role — it's a character. The README defines five design pillars:

- **Strong Personality** — Real character and voice, not generic templates
- **Clear Deliverables** — Concrete outputs, not vague guidance
- **Measurable Success Metrics** — "Done well" is defined upfront
- **Proven Workflows** — Step-by-step processes tested in production
- **Learning Memory** — Pattern recognition built into the agent's context

Some of the agent voices are genuinely memorable:

- **Evidence Collector:** *"I don't just test your code. I default to finding 3–5 issues and require visual proof for everything."*
- **Reddit Community Builder:** *"You're not marketing on Reddit you're becoming a valued community member who happens to represent a brand."*
- **Whimsy Injector:** *"Every playful element must serve a functional or emotional purpose. Design delight that enhances rather than distracts."*

---

## Getting Started in Five Minutes

If you're using Claude Code, setup is one command:
```bash
# Clone the repo
git clone https://github.com/msitarzewski/agency-agents

# Copy agents to your Claude Code directory
cp -r agency-agents/* ~/.claude/agents/

# Then activate any agent in your sessions:
# "Hey Claude, activate Frontend Developer mode..."
```

For other tools, the repo ships conversion and install scripts that auto-detect what you have installed. Supported platforms include:

**Claude Code · GitHub Copilot · Cursor · Aider · Windsurf · Gemini CLI · OpenCode · Qwen Code · Antigravity · OpenClaw**

---

## Real-World Multi-Agent Scenarios

The power isn't one agent in isolation — it's assembling the right team for a mission.

![Blog image](https://ik.imagekit.io/wurk/Startup_strategies_and_marketing_tips_CFUpxumdO.png)

---

## Why This Matters More Than It Looks

There's a temptation to dismiss agency-agents as "just prompts." That would be a mistake.

When a general-purpose AI tries to do everything, it regresses toward average. It writes marketing copy that's technically correct but lacks platform instincts. It reviews code without the focused callousness a good reviewer needs. Specialization creates better outputs in human organizations, we've always known this. agency-agents applies that same logic to AI.

The agents aren't more capable in raw terms but they are **more intentional**. The Frontend Developer knows that Core Web Vitals matter. The Deal Strategist knows a deal without a champion is a wish not a pipeline item. That intentionality baked into the agent's identity is where the value lives.

There's also something worth noting about the **community-driven nature** of this project. With 43+ contributors and translations into Simplified Chinese, agency-agents has become a living resource. Someone with real expertise in Korean business culture built that Korean Business Navigator. Someone who knew the French consulting market built that specialist. This is knowledge generic AI training often flattens and the community is encoding it properly.

---

## A Few Honest Caveats

**Agents are only as good as their prompts.** Quality varies across the roster. The core engineering agents are clearly battle-tested; newer community contributions may need refinement for your context. Read the file before relying heavily on any agent.

**Context switching has a cost.** Having 147 agents available doesn't mean using 147 on one project. Be intentional about when and why you switch modes.

**It's structure, not magic.** These agents don't give the AI new capabilities they give it a more constrained focused context. If the underlying model can't do something neither can the agent. What they eliminate is the overhead of resetting that context yourself every time.

---

## Final Thought: The Agency Model Is the Future

The shift from "one AI that does everything" to "a team of specialized agents" mirrors a maturity curve we've seen in every technology cycle. We started with one computer running one program. Then multitasking. Then specialized processors for graphics, networking, security. AI is on the same path.

agency-agents is early on that curve it's prompt files, not a sophisticated orchestration platform. But the philosophy is sound, the execution is thorough for an open-source project and 60,000 stars suggest the idea is resonating widely.

If you work with Claude Code, Cursor or any of the supported tools regularly. It's worth thirty minutes exploring this repo. You may not use all 147 agents but the ones that fit your workflow might genuinely change how you work with AI.

**→ [github.com/msitarzewski/agency-agents](https://github.com/msitarzewski/agency-agents)**