The Problem This Solves

If you have a library of AI agent skills — or any kind of digital product that you create and release regularly — you know the problem. Every time you launch a new skill, you need to do the same set of tasks: write the announcement email, create the social media posts, set up the shopping cart product, post to the community, update the spreadsheet. Manually. Every time.

James Maduk from TrainingSites.io has 165 skills in his library. Launching each one manually isn’t a workflow — it’s a full-time job. So he built a different system: an AI department manager with four employees working in parallel.

The Department Manager Model

Here’s how the system is structured:

• The Manager — A top-level Claude Cowork agent that receives the task, delegates to employees, and monitors results
• Employee 1: Email Marketing — Writes the email launch sequence for the skill
• Employee 2: Social Media — Creates community posts and social content
• Employee 3: Shopping Cart — Builds the FluentCart product listing and description
• Employee 4: Course Creator — Builds a course in FluentCommunity with lessons based on the skill

All four employees work in parallel — not one after another, but simultaneously. The manager delegates, and four agents go to work at the same time.

“This is not one, two, three, four. It is one and then four going down. There’s four agents working in parallel on different tasks. It’s four employees. It’s not one employee going from step to step to step.” — James Maduk

How the Launch Happens

The trigger is simple. James tells the manager: “Lesson Hook Creator.” That’s it. The manager reads the skill metadata, understands what the skill does, and launches the pipeline.

Here’s what the four employees produce:

• Email launch series — Three emails drafted and ready for FluentCRM
• Community feed post — Published directly to FluentCommunity
• Social media posts — Generated for multiple platforms
• FluentCommunity course — Created with sections and lessons written for the skill
• Zip package + spreadsheet update — The skill is packaged for download and the library spreadsheet is updated automatically
• FluentCart product — The shopping cart listing is set up (or saved as a file for manual entry if the API doesn’t support it)

All of that from a single two-word instruction to the manager.

What the Live Run Looked Like

James ran this live on camera for the Lesson Hook Creator skill. Here’s what happened:

1. He typed “lesson hook creator” to the department manager
2. Four parallel agents activated immediately (visible in the Claude Cowork sidebar)
3. The agents ran for approximately 35-43 steps each across different tasks
4. One agent hit a known issue (FluentCommunity course creation via API) and found an alternative approach — then continued
5. The FluentCart product description couldn’t be written to the cart directly (API limitation), so the agent saved it as an HTML file and flagged it for manual entry
6. Everything else completed: course created, emails drafted, community post published, social posts generated, zip packaged, spreadsheet updated

✓ Check Your Work: In any agentic workflow like this, expect some steps to hit API limitations or errors. That’s normal. The key is that the agent flags the issue, saves its work, and moves on rather than stopping the whole pipeline.

The Human-in-the-Loop Layer

All of the content the agents produce is in draft mode by default. Nothing goes live without James’s review. This is intentional — and it’s how James recommends running any agentic system that touches public-facing content.

The workflow is: agents do the work, James reviews and approves, content goes live. The agents handle the time-consuming production work. James handles judgment and quality control.

“It’s my company. I’m going to approve it. So that work is done.” — James Maduk

The Scalability Implication

The most significant part of this demonstration isn’t what the agents did — it’s the math. If one skill launch takes the agents about 35-45 minutes to complete in parallel, and James has 165 skills in his library, the theoretical time to launch the entire library is the same as launching one skill: the parallel runs happen simultaneously.

That’s a different relationship with scale than most educators have ever experienced. The constraint is no longer time or bandwidth — it’s judgment. What skills do you want to launch? What pricing? What courses do you want to build? Those decisions still belong to the human. The production work doesn’t.

How to Build This for Your Skill Library

1. Start with your library — Document your skills in a spreadsheet with metadata (name, description, price, category)
2. Define your department — What are the jobs that need to happen every time you launch a skill? Email? Social? Cart? Course?
3. Build the manager prompt — Give the manager clear instructions on what each employee is responsible for and what output looks like
4. Connect your tools — The manager needs MCP access to FluentCRM, FluentCommunity, FluentCart, and your social platforms
5. Run a test launch — Pick one skill, run the department, review what comes back
6. Refine and repeat — Adjust employee instructions based on what the first run produces

The Bottom Line

This is what unlimited bandwidth looks like in practice. One manager, four parallel employees, one command, full skill launch in under an hour. If you’re building an AI skills library — or any kind of product catalog that you release regularly — the department model is the system that makes it sustainable to keep building and shipping.

The agents do the production work. You do the work that only you can do.