150 AI Specialists You Can Hire in 30 Seconds

Hiring used to take months. Write a job description, post it to LinkedIn, screen applicants, run interviews, negotiate, onboard. Six weeks minimum. For a growing product team trying to ship, that timeline is brutal.

Now compare that to installing an AI agent. Copy a prompt. Paste it into your Claude Code configuration. Ask the agent a question. Total elapsed time: under a minute.

We recently launched the Agents channel on AI Skill Market, featuring 150 specialist agents drawn from the MIT-licensed msitarzewski/agency-agents repository. Each agent is a carefully crafted persona with deep domain expertise, ready to drop into any Claude-compatible workflow.

Key Takeaways

• 150 agents span 15 categories including engineering, marketing, design, sales, and spatial computing
• Every agent is MIT licensed and attributable to msitarzewski/agency-agents
• Installation takes under 30 seconds on Claude Code, Cursor, Copilot, and other platforms
• Agents are persona-based prompts, not standalone tools, so they compose with any MCP server or skill
• You can run an entire virtual agency without hiring a single human specialist

What is the agency-agents library?

The agency-agents project is a curated collection of persona prompts designed to turn a general-purpose large language model into a focused specialist. Each agent includes a role definition, core competencies, tone guidance, decision frameworks, and examples of the kinds of questions it answers well.

Think of it as a staffing manual for the AI era. Instead of one giant assistant that does everything mediocre, you get 150 narrow experts who each nail their lane.

The 15 categories break down like this:

• Engineering (26 agents) — frontend, backend, DevOps, security, data engineering, and more
• Marketing (29 agents) — SEO, content, growth, email, brand
• Design — UI, UX, brand guardians, illustration
• Sales — discovery, qualification, closing, retention
• Testing — unit, integration, E2E, accessibility
• Product — PM, researcher, analyst, strategist
• Project Management — scrum master, program manager, operations
• Paid Media — Meta, Google, TikTok, LinkedIn
• Support — customer success, tier 1, tier 2, documentation
• Specialized — orchestrator, researcher, blockchain auditor
• Academic — grad student helper, citation manager, thesis advisor
• Game Development — solo dev, engine specialist, level designer
• Spatial Computing — AR, VR, WebXR
• Strategy — business model, pricing, competitive analysis
• Integrations — API design, webhook plumbing, third-party systems

Why personas beat raw prompting

If you've been prompting LLMs for any length of time, you probably already know that "You are an expert X" at the top of a message materially improves output quality. Agency agents take that observation seriously and productize it. Each persona is tuned over dozens of iterations to elicit a consistent voice, depth of analysis, and set of concerns specific to the role.

Compare asking "review my React code" to invoking the Frontend Developer agent, which knows to check accessibility, bundle size, hydration mismatches, render performance, and state management patterns before it even looks at your diff. The difference is night and day. We dig deeper into this contrast in How Agents Beat Prompt Engineering in 2026.

How fast is "30 seconds"?

For Claude Code, installation is literally three commands: clone the repo, symlink the agent file into your .claude/agents/ directory, and restart the session. If you use our bundled installer it drops to one command. Cursor and Copilot have slightly different flows, covered in How to Install Agents Across 8 AI Platforms.

The key insight is that agents are text files. Not Docker images, not Python packages, not npm modules with peer dependency conflicts. Text. That makes them trivially portable across any tool that accepts system prompts.

Who should use agency-agents?

Solo founders and indie hackers

If you're a solo founder, agents are the closest thing to a free cofounder. The Growth Hacker agent alone can replace a $120k/year marketing hire for early-stage experimentation. The Product Manager agent will draft PRDs that would take you a day in half an hour.

Small agencies

Running an agency is all about leverage. Every hour your humans spend on tactical work is an hour they're not spending on strategy or client relationships. Agents can take over brief generation, competitive analysis, first-draft copy, QA test plans, and dozens of other activities. Running an AI Agency With 151 Agents walks through a real workflow.

In-house teams

For in-house teams, agents serve as institutional knowledge amplifiers. Pair your Security Engineer agent with your existing codebase and you get a 24/7 security reviewer that never gets tired and never forgets to check for prototype pollution in that one weird legacy file.

What about cost?

Agents themselves are free — they're just prompts. Your cost is whatever you're already paying for Claude, Cursor, Copilot, or your LLM of choice. For heavy use, consider API routing through Claude Code with a Pro plan.

A realistic estimate: most agents complete a task in 2,000–8,000 tokens of output. At Sonnet rates, that's pennies per invocation. Even a team running 500 agent tasks per day lands in the low double digits of dollars.

Frequently Asked Questions

Can I use agency-agents commercially?

Yes. The repository is MIT licensed, which permits commercial use, modification, and redistribution with attribution. Keep the copyright notice intact and you're covered.

Do agents work with models other than Claude?

Mostly, yes. The prompts are written in plain English and use techniques that generalize across frontier models. Some agents lean on Claude-specific behaviors like tool use and XML tagging, which can degrade on GPT or Gemini.

How do I pick the right agent for a task?

Start with the category. If your problem is a landing page, you're in Marketing or Design. If it's a bug in production, it's Engineering. Read three or four agent descriptions in the relevant category and pick the one whose scope matches your actual problem.

Can agents call other agents?

Yes, but you need an orchestrator. The Agents Orchestrator specialized agent is purpose-built for this, and there are also MCP-based patterns documented in our MCP guide.

Is there a way to submit my own agent?

Absolutely. Fork the upstream repo and send a PR to msitarzewski/agency-agents, or publish directly to AI Skill Market via /submit.

What's next?

Over the next few weeks we'll publish deep dives into individual agents, category roundups, and platform-specific install guides. Bookmark the Agents channel and check back weekly.

The era of hiring humans for every role is ending. The era of assembling a virtual team in 30 seconds has begun. The question isn't whether you'll use AI agents — it's whether you'll get there before your competitors.

Browse all 150 agents at aiskill.market/agents or submit your own skill.

Sources

• msitarzewski/agency-agents on GitHub — the MIT-licensed source repo
• Claude Code documentation
• Anthropic's GitHub organization
• AI Skill Market Agents channel