Session Overview
This Campus VIP session covered two major themes: YouTube channel management at scale using API automation, and the complete TrainingSites AI Operating System — a six-piece framework that turns Claude into a persistent, self-improving business operator. The session included live demos of both systems and practical strategies for community gamification.
Date: Monday, March 24, 2026 | Duration: 1h 41m | Participants: James, Boki
The AI Operating System — Six Connected Pieces
The AI Operating System is a project folder structure that gives Claude persistent memory, organizational awareness, and the ability to route work through specialized departments. Here are the six pieces:
1. Org Chart
Define your business departments as AI roles. TrainingSites uses: Chief of Staff (Dean), Community Manager, Education Manager, Sales Manager, and Marketing Manager. Each department has its own CLAUDE.md file with role boundaries, recurring tasks, and standard operating procedures. When you tell Dean to create a course, Dean knows to route it through the Education Manager, which already has the skills and context to execute.
2. Memory System
At the start of every session, the agent reads shared context files (ICP, brand, offers, systems) plus department-specific memory. At the end of every session, it writes back what was learned — corrections, preferences, decisions. Only the 8 most recent entries are kept in active memory to manage token usage, but all history is preserved in the agent logger for deeper lookups.
3. Agent Logger
A custom WordPress plugin that records every skill execution, task completion, time saved, and estimated dollar value. This creates an audit trail: when did what agent do what task? The logger feeds into both the operations board and the Pinecone vector database, ensuring nothing gets lost between sessions.
4. Pinecone Vector Database
Pinecone is a vector database (free tier available at pinecone.io) that stores AI-readable versions of all WordPress content. Using the AI Engine WordPress plugin, content auto-syncs to Pinecone every 5 minutes. This powers: the site chat agent (checks page → Pinecone → web), cross-session knowledge retrieval, and potential future API access for selling knowledge base access.
5. Operations Board
A FluentBoards task board with five columns: Ideas, Waiting on James, In Progress, Blocked, and Done This Week. Every morning, Dean checks the board for pending work. Ideas get moved to In Progress automatically; blocked items surface for human decision-making. This replaces the need to manually assign tasks each session.
6. Scheduled Tasks
Automated daily and weekly jobs configured in Claude’s scheduled tasks: morning intelligence report, campus ambassador community sweeps (morning + evening), partnership scans, weekly newsletter assembly, and topical authority content generation (25-50 articles per day from the TAM’s 7,500 questions).
YouTube Channel Management via API
Boki demonstrated a workflow for managing YouTube channels at scale using Claude Code and the YouTube Data API:
Batch Description Optimization
Connect the YouTube Data API through Google Cloud Console. Pull all video metadata including titles, descriptions, view counts, and existing articles. Have Claude generate optimized descriptions with: keyword-rich summaries, timestamp markers, related video links, subscribe CTAs, and article cross-links. Batch update up to 50 videos at a time within Google’s 10,000 daily API quota.
Automated Comment Management
Pull top comments per video using the API. Have Claude draft personalized replies following brand voice rules. Review drafts, then batch-post approved replies. Especially useful for replying to old comments on evergreen videos — viewers may re-engage when they see a late reply.
Content Gap Analysis
Cross-match all YouTube videos with website articles. Identify which videos lack companion articles. Claude recommends the top 3 articles to write based on video performance and content gaps. Also performs SEO audits with automated scripts for Google Search Console indexing.
Community Gamification Strategy
Boki shared a gamification approach for community-led course gating designed to drive engagement before monetization:
The first two course modules are free for all community members. To unlock Module 3 and beyond, members complete simple engagement milestones: upload a profile photo, write a bio, make 5 posts, and comment on others’ content. Thresholds are intentionally low — completable within a single day of active participation.
Between the free and paid tiers, offer a complimentary 1-on-1 session as a reward for completing the free modules. This doubles as a discovery call. Record the session, use AI to generate a personalized business plan, and send it as a follow-up. Even non-buyers walk away impressed, generating word-of-mouth.
Cowork vs Claude Code — When to Use Each
Use Cowork for content waterfalls (repurposing videos, creating emails, posting to community), running skills that interact with MCP connectors, and any task where the output goes directly to WordPress, FluentCRM, or social media.
Use Claude Code for planning and building (plugins, skills, project structures), code-heavy work, and when you need the project context to persist across sessions with git-level tracking.
GoHighLevel MCP Limitations
Boki reported that the GoHighLevel API via MCP is limited in several ways: visual builders (email workflows, automations) cannot be controlled via API; the MCP connection is slow and sometimes hangs even with proper permissions; community content management works for basic posts but struggles with structural changes. Blog posts and basic content management work, but anything involving drag-and-drop builders does not.
Key Quotes
“I don’t do skills anymore. All I do is tell Dean what I want done.” — Once the operating system is set up with departments and SOPs, you stop running individual skills. You describe outcomes and the system routes execution.
“You should really focus on getting people first. Even if you have a shell, just build a shell.” — Don’t spend months perfecting a community nobody has joined. Build the minimum, get people in, iterate as you go.
“You’ve got to find a way to capture the work that Claude is doing and organize it.” — Without logging and memory, every session starts from scratch. The agent logger → Pinecone pipeline creates institutional knowledge.
Tools and Resources
AI Tools: Claude Code, Claude Cowork, Claude Desktop, NotebookLM, Remotion, AI Engine (WordPress plugin)
Platforms: WordPress, FluentCommunity, FluentCRM, FluentBoards, GoHighLevel, Hostinger VPS
Databases: Pinecone (free vector database at pinecone.io), WordPress Agent Logger plugin
APIs: YouTube Data API, Google Search Console, Google My Business, Google Sheets, Google Drive
Courses Referenced: AI BUILD: Meet Your AI Team, AI BUILD: Train Your First AI Agent Employee