Irv Cassio • AI Enthusiasts Group • March 20, 2026
02 — VENU POLAKI
The Easy Button for Enterprise IT
What if solving complex IT problems was as simple as pressing a button? AI Skills and MCP connections make this possible — turning multi-step manual processes into one-command solutions.
🔎
Diagnose
AI reads the ticket, pulls logs, checks systems — all in seconds
⚡
Resolve
Identifies root cause and suggests fixes based on historical patterns
📚
Learn
Captures knowledge from every incident, making the team smarter over time
“A skill solving our problems is like having an Easy Button.”
03 — VENU POLAKI
What is MCP?
Model Context Protocol (MCP) is an open-source standard created by Anthropic that lets AI models securely connect to external data sources, tools, and APIs. Think of it as USB-C for AI — one universal connector instead of custom integrations for every system.
📂 Filesystem MCP
AI can read, write, and search files on your computer. “Read the CSV on my desktop and summarize it.”
🐙 GitHub MCP
AI can create PRs, search code, manage issues. “Create a PR for the bug fix on branch main.”
🗃 Database MCP
AI queries Postgres, MySQL, Oracle directly. “Show me all orders from the last 7 days with status pending.”
📧 Email MCP
AI reads and sends email through Apple Mail, Outlook, etc. “Find emails from Sarah about the Q1 report.”
MCP servers are always-on connectors that run as background processes. The AI calls them whenever it needs to interact with external systems.
04 — VENU POLAKI
What are Claude Skills?
Claude Skills are reusable, custom instruction sets — folders of text, code, and data — that teach Claude to perform specific tasks consistently. They turn a general-purpose AI into a specialist for your needs without re-explaining instructions every time.
🎫 /triage
Say “/triage 45678” and Claude looks up the ticket, checks related systems, and provides a diagnosis — all automatically.
📈 /storelookup
Say “/storelookup 4521” and Claude returns store details, device inventory, GPS coordinates, and open tickets.
🌅 /endofday
At end of day, Claude reviews your work, creates knowledge docs, and updates the team knowledge base automatically.
📸 /screenshot
Claude reads your latest screenshot, analyzes the content, and takes action — like reading error messages and suggesting fixes.
Skills are on-demand instructions invoked with slash commands. They load context dynamically and can chain together multiple MCP tools.
05 — VENU POLAKI
MCP vs Skills: What’s the Difference?
This is a common point of confusion. Here’s the simplest way to think about it:
🔌 MCP Server
The connection
✦ Always-on background process
✦ Connects to one external system
✦ Provides raw tools (query, read, write)
✦ Like a USB cable to a printer
✦ Configured in .mcp.json
⚡ Claude Skill
The workflow
✦ On-demand, invoked by slash command
✦ Orchestrates multiple tools & MCPs
✦ Contains domain knowledge & logic
✦ Like a print macro that formats & prints
✦ Lives in ~/.claude/skills/
Together: A ticket system MCP gives Claude the ability to read tickets. A triage skill tells Claude how to diagnose the ticket, which systems to check, and how to format the results.
06 — VENU POLAKI
MCP Architecture
How AI clients connect to enterprise systems through MCP:
Claude Codeor Claude Desktop
→
Your Machinelocalhost
→
MCP: Oracle SQL
MCP: New Relic
MCP: Filesystem
→
Oracle DB
New Relic API
Local Files
Each MCP server is a lightweight process on your machine. The AI client discovers available tools at startup and calls them as needed. All communication stays local — your data never leaves your network.
07 — VENU POLAKI
Building Your Own MCP Server
Any developer can create an MCP server. The protocol is open-source and well-documented.
---name: triage
description: AI-powered ticket
diagnostics and system checks
trigger: "triage ticket",
"diagnose incident"
---
# Triage Skill
When invoked with a ticket ID:
1. Look up ticket via HDA MCP
2. Parse the issue description
3. Check New Relic for anomalies
4. Check Splunk for related logs
5. Cross-reference known issues
6. Provide diagnosis + next steps
11 — VENU POLAKI
Available Skills Library
Built-in Skills
docxCreate, read, edit Word documents
pdfRead, merge, split, OCR PDFs
pptxCreate and parse PowerPoint decks
xlsxOpen, read, edit spreadsheets
frontend-designProduction-grade web UI components
presentHTML slide deck generator
claude-apiBuild apps with Claude API
Template Skills
skill-creatorCreate & optimize skills
mcp-builderBuild MCP server integrations
brand-guidelinesBrand consistency tooling
internal-commsInternal communication drafting
web-artifactsBuild web-based artifacts
12 — VENU POLAKI
EASY The Triage Skill: Instant Ticket Diagnosis
One command replaces a manual process that takes engineers through multiple systems and screens.
❌ Before (Manual)
Log into ticket system
Search for ticket number
Read through pages of detail
Open New Relic in another tab
Search for related app
Check Splunk for logs
Cross-reference manually
Form a hypothesis
⏱ 15–30 minutes per ticket
✅ After (Triage Skill)
> /triage 45678
Ticket #45678 — Store Terminal OfflineStore: Location 5-8852, Dublin
Reported: 2 hours agoDiagnosis:Network switch showing intermittent
packet loss (12%) since 14:30 UTC.
Correlates with New Relic alert #NR-7821.Recommended Action:Restart network switch port 24.
⏱ 30 seconds, fully automated
13 — VENU POLAKI
The ROI: Even Small Savings Compound
Just automating the ticket lookup step — before any diagnosis — already saves significant time.
📐 The Calculation
Ticket lookups per day2
Time per manual lookup~2 min
Working days per year260
Time saved per person/year~17 hours
Team of 20 engineers~340 hours
340 hrs/year
That’s 8.5 work weeks returned to the team — from one simple automation
Beyond Time Reduced context switching • Fewer login frustrations • Instant access to structured data
Conservative estimate — login (15s) + search (30s) + navigate (15s) + scroll/read (60s) = ~2 min per lookup
14 — VENU POLAKI
EASY Service Resolution Toolkit (SRT)
The Easy Button for Major Incidents
When a major incident impacts hundreds of users, the SRT skill scans multiple systems simultaneously to identify the root cause — what would take a team of engineers hours of manual investigation.
🔎
Scan
Queries New Relic, Splunk, databases, and monitoring tools simultaneously
📈
Correlate
Cross-references performance trends, error spikes, and infrastructure changes
🎯
Pinpoint
Identifies root cause and provides actionable remediation steps
Cutting diagnosis time from 45 minutes to 5 minutes during a major incident doesn’t just save engineer time — it directly protects revenue.
15 — VENU POLAKI
Language is No Barrier
LLMs don’t care what language a ticket is written in. The triage skill translates the full detail — not just a partial summary you might get from a person in the middle.
🇮🇹 Ticket in Italian
“Il terminale nel negozio di Milano non riesce a completare le transazioni. L’errore si verifica dopo le 15:00 e sembra essere correlato alla connessione di rete...”
🌐 Triage Output (English)
Summary: Store terminal in Milan failing transactions after 3 PM. Network connectivity appears to be the root cause.
Full Context: Complete ticket detail preserved, not summarized. All technical specifics, timestamps, and error codes translated accurately.
🇫🇷 French
🇩🇪 German
🇪🇸 Spanish
🇵🇹 Portuguese
🇹🇷 Turkish
🇨🇳 Chinese
🇯🇵 Japanese
16 — IRV CASSIO
AI-Powered Device Validation
Claude analyzed 47 tablet screenshots across 10 devices and compiled a comprehensive validation report — including issue analysis and action items.
Report Output
Devices Tested10
✅ Fully Updated5
⚠ Partially Updated2
❌ Update Failed3
Screenshots Analyzed47
What Claude Did
Read 47 screenshots via vision
Identified device serial numbers
Checked firmware versions
Compared against target versions
Classified update status
Analyzed error messages
Identified root causes
Generated structured report
Created action items per store
Manual: ~3 hours → With AI: ~5 minutes
17 — IRV CASSIO
Claude in Chrome: The AI Negotiator
A user let Claude’s Chrome extension take over an AT&T support chat. The result? A $100 billing credit — fully autonomously.
How It Worked
User opened AT&T support chat
Told Claude: “I need a refund for the outage”
Claude took over the chat entirely
Escalated politely, cited service terms
Handled repetitive responses patiently
Secured a $100+ credit
Also prepared an FCC complaint as backup
Why AI Wins Here
✦ Never loses patience with repetitive responses
✦ Stays on-message through long negotiations
✦ No emotional concessions under pressure
✦ Can prepare legal escalation simultaneously
Enterprise Parallels
The same Claude Chrome extension can interact with vendor portals, internal dashboards, and web-based tools — extending AI to any browser-based system.
Source: Nate B. Jones, March 2026
18 — VENU POLAKI & JOE SIEGMANN
The End-of-Day Knowledge Skill
Inspired by Joe Siegmann’s idea — AI reviews your day’s work and auto-generates knowledge that makes skills smarter and humans more informed.
📝
Review
Claude reviews all work performed: tickets resolved, code committed, issues investigated
🧠
Capture
Creates structured markdown knowledge files: patterns found, solutions that worked
🔄
Compound
Skills continuously learn. Tomorrow’s triage is smarter because of today’s resolutions.
The Flywheel Effect: Every incident resolved creates knowledge → knowledge improves skills → skills resolve incidents faster → engineers have more time for complex work → more knowledge captured. The system gets smarter every day.
19 — IRV CASSIO
Learn & Get Certified
Anthropic launched free courses and its first technical certification in March 2026.
🎓 Anthropic Academy
Free, self-paced courses with certificates. 15 courses across three tracks:
Developer:Claude Code in Action • Claude API • Intro to MCP • MCP Advanced • Agent Skills • Subagents
Product:Claude 101 • Cowork • Bedrock • Vertex AI
AI Fluency:Framework & Foundations • Educators • Students • Nonprofits
🏆 Claude Certified Architect (CCA)
Anthropic’s first official technical certification — proctored, architecture-level exam.
✦ 60 proctored questions
✦ No external tools or AI allowed
✦ Scored on 1,000-point scale
✦ Free for first 5,000 partner employees
5 Competency Domains
1. Agentic Architecture (27%)
2. Claude Code Workflows (20%)
3. Prompt Engineering (20%)
4. API & Tool Use (18%)
5. Safety & Governance (15%)
20 — IRV CASSIO
NVIDIA NemoClaw & OpenClaw
Announced at GTC 2026, NVIDIA’s NemoClaw brings enterprise-grade security to autonomous AI agents — what Jensen Huang calls “the new Linux.”
🌐 OpenClaw
The fastest-growing open-source project in history (247K+ GitHub stars). An open platform for autonomous AI agents that can self-evolve, learn, and take action.
✦ Open-source agent framework ✦ Community-driven development ✦ “Every company needs a strategy for it” — Jensen Huang
🔒 NemoClaw (Enterprise)
NVIDIA’s enterprise layer on top of OpenClaw. Adds sandboxed environments, policy-based guardrails for network access, data privacy, and security.
✦ One-command install via OpenShell ✦ Privacy & security guardrails built in ✦ Runs on DGX Station, DGX Spark, or any GPU
What This Means: Imagine always-on AI agents running locally in your data center — monitoring systems, triaging incidents, and resolving issues 24/7 with enterprise-grade security. The skills and MCP patterns we’re building today are the foundation for this future.
21
Next Steps
Key priorities for the upcoming quarter:
1 Reduce Costs
Explore local LLM options to reduce API costs. Evaluate GitHub Copilot licensing for developer productivity gains.
2 Role-Based Access
Implement proper role-based access controls to meet security and compliance requirements across all AI tools.
3 Training & Testing
Schedule more hands-on training sessions for the team. Establish thorough testing cycles before production rollouts.
4 Expand Skill Library
Identify and develop additional use cases. Expand skill library and knowledge base for broader coverage.