ai-presentations/week1-mcp-skills/presentation.md

The AI Multiplier

MCP + Skills in Action

The Copy-Paste Loop

Sound familiar?

sequenceDiagram
    You->>LLM: I'm 10 minutes away from a demo, getting this error        
    LLM->>You: Try this...
    You->>LLM: That didn't work! New error!
    LLM->>You: Gather this info..
    You->>LLM: Here it is...
    LLM->>You: Try this...
    loop 
    You->>LLM:😤
    LLM->>You:😭
    end

You are the middleware. You copy. You paste. You translate.

The AI is blind, deaf, and handless. You're doing all the work.

MCP: Giving AI Hands

Model Context Protocol — open standard, created by Anthropic (2024)

One sentence: MCP gives tools to AI.

sequenceDiagram
    You->>LLM: I'm 10 minutes away from a demo, getting this error        
    LLM->>You: Let me check that..
    LLM->>OCP: Get pod list
    LLM->>OCP: Get pod logs
    LLM->>ArgoCD: Get GitOps application
    LLM->>Git: Get Git repo
    LLM->>LLM: Plan fix
    LLM->>You: Here is the plan. Requesting Approval
    You->>LLM: Approved
    LLM->>Git: Push fix to Git
    LLM->>OCP: Verify change applied
    LLM->>You: Applied Successfully ✅

You approve the plan. AI does the legwork.

What Can MCP Connect To?

Anything with an API. Today, in this room, we have:

• OpenShift / Kubernetes — full cluster operations

• Git (Gitea, GitHub, GitLab) — repos, commits, PRs

• ArgoCD — GitOps application management

• SecureTransport — admin API for file transfer flows

• Databases, monitoring, CI/CD, ticketing...

There are 10,000+ MCP servers available. And you can build your own.

Skills: Giving AI Expertise

MCP gives AI tools. Skills give AI knowledge of how to use them well.

graph LR
    subgraph WITHOUT["❌ Without Skills"]
        direction TB
        BP["System Prompt<br/>20,000+ tokens<br/>Everything, always"]
    end

    subgraph WITH["✅ With Skills"]
        direction TB
        AP["System Prompt<br/>~500 tokens"]
        AP -->|"task matches"| L1["Load: Healthcheck<br/>+2,000 tokens"]
    end

    style WITHOUT fill:#ffcccc,stroke:#cc0000
    style WITH fill:#ccffcc,stroke:#00cc00
    style BP fill:#ff6b6b,stroke:#333,color:#fff
    style AP fill:#51cf66,stroke:#333,color:#fff
    style L1 fill:#4a9eff,stroke:#333,color:#fff

Lazy-loaded expertise. Scan descriptions → load on demand → unload when done.

A Skill is an SOP

Without Skill 🎲

"Check if my server is secure"

• Runs random checks from training data
• Misses half the important ones
• Suggests changes that break your infra

With Skill 📋

"Check if my server is secure"

• Loads healthcheck SKILL.md
• Follows structured audit checklist
• Firewall → SSH → Updates → Services
• Prioritised report with specific fixes

Your best engineer writes it once. Every agent benefits, forever.

People leave. Skills persist.

Tools + Skills + Identity

graph LR
    T["🔧 Tools (MCP)<br/>What can I do?"] --> A["🤖 Agent"]
    SP["🧠 Identity (System Prompt)<br/>Who am I?"] --> A
    SK["🎯 Skills<br/>How do I do X well?"] --> A
    
    style T fill:#ff6b6b,stroke:#333,color:#fff
    style SP fill:#ffd93d,stroke:#333,color:#000
    style SK fill:#4a9eff,stroke:#333,color:#fff
    style A fill:#51cf66,stroke:#333,color:#fff

A workshop full of power tools, no training? Dangerous.

Power tools + SOPs + a qualified operator? Productive.

DEMO

Watch the AI. Watch me. Neither of us needs the other.

What You Didn't See

• Every command the AI ran (and there were dozens)

• Every error it hit and recovered from

• Every tool it chose — and the ones it considered and rejected

• The context it carried across steps without being reminded

That's the loop flip.

You used to do all of that. Now you approve the plan.

The 3x Multiplier

1x

You, working.

The baseline.

2x

You work + AI works.

Parallel output.

3x

Parallel work + AI adds value you didn't ask for.

Compounded output.

The person leveraging AI will outcompete the person who isn't.

This isn't a threat. It's an opportunity you're choosing to take — or not.

Next Week

You just saw what's possible.

Next Friday: how to get this for yourself.

CLAUDE.md, system prompts, project context — the practical playbook for making AI actually understand your work.

🎯