← Back to the workshop

Core AI Concepts

Seven building blocks will help you understand what AI can do, what information it needs, and where you stay in control.

If you're already familiar with these concepts, jump to Hilary's guidance on choosing a tool or privacy and permissions.

01 Setup

🧠 Model

The brain

The AI that interprets your request and produces a response. The same model can appear in different apps, but its tools, instructions, and available context affect what it can do there.

Examples
  • Claude models
  • GPT models
  • Gemini
  • Llama
You might also hear
  • LLM
  • Large language model
  • Foundation model
  • Frontier model
  • "the AI"
02 Setup

📱 App

The body / wrapper

The product 'wrapped' around the model. You might chat in a browser (Claude.ai) or code in your terminal (Claude Code), but the brain inside can be the same.

Examples
  • Claude.ai
  • Cowork
  • Claude Code
  • ChatGPT / Codex
  • Cursor
You might also hear
  • Client
  • Interface
  • Product
  • Surface
  • Wrapper
  • Harness
  • IDE (Integrated Development Environment)
03 Setup

🌐 Environment

Where it can play

Where the agent's tools run: on your computer or on a remote computer. Some apps let you choose. A local working folder does not mean the model runs offline or that its file access is unlimited.

Examples
  • A remote workspace managed by the provider
  • An isolated workspace with selected files connected
  • A local folder, subject to your app’s permissions
You might also hear
  • Sandbox
  • Runtime
  • Host
  • Workspace
  • Container
  • Local vs. Cloud
04 Operating

📝 Instructions

What you tell it to do

Anything you write down once so you don't have to say it every time.

Examples
  • You talking to AI
  • A saved 'weekly update' prompt
  • A Markdown (.md) text file
  • A SKILL.md inside Claude Code
You might also hear
  • Skill
  • Prompt
  • System prompt
  • Custom instructions
  • Custom GPT
  • Persona
  • Project rules
05 Operating

🛠️ Tools

What actions it can take

Anything the AI does instead of just talking — searching the web, writing a file, reading your calendar, etc.

Examples
  • Web search
  • Read or write a file
  • Read your calendar
  • Send an email
  • Run code
  • Click on a web page button
You might also hear
  • Function calls
  • Tool use
  • Actions
  • Capabilities
  • Computer use
  • Browser use
06 Operating

🔗 Connections

What systems and data it can access

Other apps you want the AI to have access to: your inbox, calendar, Drive, Slack, Asana, etc.

Examples
  • Gmail
  • Google Calendar
  • Drive, Dropbox, Notion
  • Slack
  • Github
  • A folder on your computer
You might also hear
  • Integration
  • Connector
  • Extension
  • MCP
  • MCP server
  • API
  • Plugin
  • Data source
07 Operating

⚡ Triggers

What sets it in motion

How it knows to do something. Simplest trigger is you asking (chat), but you can also automate on a schedule or when something else happens.

Examples
  • You typing in chat
  • A slash command
  • Run daily at 9am
  • An email arriving
  • A push to Github
You might also hear
  • Schedule
  • Cron / Cron job
  • Automation
  • Routine
  • Workflow
  • Hook
  • Webhook
Key Takeaways
#1
Knowing what to reach for matters as much as knowing how to use it.

Tools and capabilities change constantly — what's connected to what, which app can do what this month versus last. Understanding these seven concepts is what lets you evaluate any new tool that shows up, instead of memorizing a list that'll be outdated by next year.

#2
This isn't just practice for home.

Practicing on an everyday task helps you learn how to break down work, provide context, and check results. Those habits transfer to your job too. Use your own judgment with family information, and follow your organization's rules when you apply the method at work.

↑ Back to top