AI Beyond the Basics — Your Cheat Sheet

01 — Claude's Toolkit

Cowork & Design

Two new features that move AI from a chat window into your actual workflow.

Claude Cowork

A real-time AI collaborator. Claude joins your workspace and works alongside you — seeing your edits, offering suggestions, and contributing to documents, spreadsheets, and projects as you work. Think of it as a smart teammate who's always available.

Claude Design

Create visual content through conversation. Describe a landing page, dashboard, or presentation in plain English and Claude generates polished designs. Iterate by talking — “make the header bigger,” “use a warmer palette,” “add a testimonial section.”

01

Cowork Starter

I'm working on [project]. Here's my current draft: [paste text]. Review it for clarity, suggest improvements, and help me restructure the weakest sections. Let's work through this together.click to copy

02

Design Starter

Design a professional landing page for [business/event]. Include a hero section with headline, a features section with 3 key benefits, testimonials, and a call to action. Use a clean, modern style.click to copy

02 — Extensibility

Understanding MCPs

Model Context Protocol lets AI connect to your tools — calendar, email, files, databases, and more. Think of them as plugins.

What is MCP?

An open standard that lets AI tools connect to external services. Instead of copy-pasting data into AI, MCPs let the AI reach out and get it directly. Claude pioneered this approach, and it's becoming an industry standard.

ChatGPT

Actions & Connectors

ChatGPT uses a curated marketplace of built-in connectors. Easy to set up, works out of the box. Great for standard integrations like Google, Slack, and Zapier. Less flexible for custom setups.

Claude

Model Context Protocol

Claude uses the open MCP standard. More powerful and flexible — connect to any tool that supports the protocol. Growing open-source ecosystem. Better for custom workflows and enterprise use.

MCP vs API — What's the Difference?

API: A way for developers to write code that talks to a service (e.g., pulling weather data into an app). Requires programming knowledge.
MCP: A plug-and-play connection that lets AI use a service directly — no coding needed. Think of APIs as the engine under the hood, MCPs as the steering wheel you actually use.

Easy MCPs to try: Google Drive (search and read your files), Slack (search messages and channels), GitHub (manage repositories), and Brave Search (web search with no tracking).

03 — Understanding Limits

Context Windows Explained

Every AI has a limit on how much information it can process at once. Understanding this limit helps you get better results.

Platform	Context Window	Roughly Equals
ChatGPT	~128K tokens	~300 pages
Claude	~200K tokens	~500 pages
Gemini	~1M tokens	~2,500 pages

Bigger Doesn't Mean Better

Even models with 1M+ token windows (like Claude Opus 4.7) start hallucinating more past the ~200K mark. More context means more room for AI to get confused, fill in gaps with made-up details, or lose track of instructions. Bigger windows are useful, but reliability drops well before you hit the limit.

Working Around the Limits

Break large documents into chunks and process them separately. Summarize sections first, then analyze the summaries. Be specific about what you need — don't paste everything. And when the data is truly too large, ask AI to write a script that processes it instead.

04 — Practical Example

When AI Should Write the Tool

Sometimes the best use of AI isn't processing your data — it's building a tool that processes it for you.

The Problem

You have a massive contact database in Excel. Thousands of rows with issues: ‘.con’ instead of ‘.com’, spaces in email addresses, duplicate entries, inconsistent phone formatting. Too much data to paste into any AI.

Remember: it's rows × columns. A "small" CSV with 3,000 rows and 25 columns is 75,000 cells. That modest file can consume 500K+ tokens — exceeding most context windows and well past the reliability threshold.

The Solution

Ask AI to write a script that cleans the data for you. The script runs on your computer, handles any file size, applies consistent rules, and catches every instance. AI is better at writing the tool than being the tool.

EXAMPLE PROMPT

Database Cleaning Script

Write a Python script that reads an Excel file called 'contacts.xlsx' and checks for these issues: 1) Email addresses with spaces (like 'email @gmail.com'), 2) Common domain typos (.con, .cmo, .gmal instead of .com, .com, .gmail), 3) Duplicate rows based on email address, 4) Phone numbers in inconsistent formats. Output a report showing every issue found, grouped by type, with the row number and original value.click to copy

Key insight: You don't need to know Python. You just need to describe what you want. AI writes the script, you run it. If something's not right, describe the issue and AI fixes the script.

Best Practices — Before You Run Anything

Always run scripts on a copy of your data — never the master file. If something goes wrong, your original is safe. If you're not comfortable reading code, have someone who is review it before running. AI-generated scripts are usually solid, but they can make assumptions about your data structure. Trust but verify.

05 — Automation

Getting Started with Agents

AI agents go beyond chat — they take action, use tools, and complete multi-step tasks. Here are three easy ways to start.

ChatGPT Custom GPTs

Build a Specialized Chatbot

Create a custom chatbot with your own instructions, uploaded files, and tools. Perfect for team FAQs, customer support, or any task you repeat often. Share with your team or keep private.

Claude Projects

Your AI Workspace

Create a project with persistent context, custom instructions, and uploaded documents. Claude remembers everything in the project across conversations. Great for ongoing work like content creation, research, or analysis.

Gumloop

Visual Agent Builder

Drag-and-drop interface for building AI automations. Use Gummie to describe what you want in plain English and it builds the workflow. No coding required. Connect to Gmail, Sheets, Slack, and more.

STARTER PROMPT

Agent Starter

I want to build an AI agent that [describe the task]. It should run [how often: daily/weekly/when triggered]. The input is [where the data comes from]. The output should be [what you want it to produce]. What's the simplest way to set this up?click to copy

06 — The Frontier

OpenClaw & AI Safety

AI tools are getting more powerful and more autonomous. Here's what's coming, and why security matters more than ever.

What is OpenClaw?

An open-source AI assistant that runs 24/7 on your own computer. You interact with it through WhatsApp, Telegram, or text. It manages your email, calendar, files, and tasks — and builds new capabilities on its own. 310K+ GitHub stars, fastest-growing open-source AI project, and its creator was hired by OpenAI within 4 months. This is the direction personal AI is heading.

The Security Reality

Cisco's security team called OpenClaw “a security nightmare.” China restricted government agencies from using it. Tools like this can access everything on your computer — files, emails, passwords, browsing history. The more powerful and autonomous the AI tool, the more critical it is to:

Contain what it can access — sandbox or limit permissions
Never grant full system access to any AI tool
Review what permissions you’re granting before enabling
Ask: “What’s the worst this could do?” before giving it access

Key principle: Power without containment is a liability. As AI tools get more capable, security and guardrails become more important, not less.

07 — Presentation

Review the Slides

Want to revisit something? The full presentation is available as an interactive slide deck.

AI Beyond the Basics — Full Presentation

14 slides covering Claude Cowork & Design, MCPs, context windows, database cleaning, AI agents, and AI safety. Navigate with arrow keys or swipe.

→

08 — Resources

Quick Links

Everything in one place.