Learn AI Tools.
In Plain English.

AI coding tools don't live in a browser. They live in the terminal — the text-based interface on your computer where you type commands and things happen. It looks intimidating: blank screen, blinking cursor, no buttons. But here's what you get in return: instead of clicking through 15 browser tabs and copying text between windows, you type one sentence and the AI does the work — because it has access to your computer, your files, your images, and the internet.

Before you start, you install the AI tool. For Claude Code:

Mac/Linux: curl -fsSL https://claude.ai/install.sh | bash
Windows: irm https://claude.ai/install.ps1 | iex

Then create a practice folder: mkdir ~/playground && cd ~/playground

Pricing: Max plan ($100–200/mo) for daily business use. Pro plan ($20/mo) for getting started. If you'll use it daily, go Max — the ROI shows up in week one.

This is where AI stops being a chatbot and starts being useful. The AI reads and edits files on your computer — with your permission. Not "paste your document into a chat window." It sees your actual files. Your proposals, spreadsheets, client folders. And it edits them directly.

The AI sees images, screenshots, PDFs, diagrams, photos of whiteboards, mockups, receipts. Drop a file in and it reads it. Take a screenshot of an error message and paste it. Drag in a PDF invoice. It reads it all and works with it like text.

This is the paradigm shift. The AI doesn't chat. It acts. It reads files, edits documents, runs commands, searches your folders, fetches web pages, connects to your apps. Each action is a "tool." You see them happening in real time as it works.

Example: you ask it to find invoices. It uses the search tool to scan every file. Then the read tool to open each match. Then the write tool to create a summary document. You see every step.

The #1 mistake: vague prompts. "Help me with my marketing" = garbage in, garbage out.

"Write a 3-email welcome sequence for my dog walking business. Casual tone, mention our GPS tracking feature, each email under 200 words, include a CTA to book a free walk" = RESULTS.

Be specific. Say what you want AND what you don't want. Give constraints. Name files. Tag specific files with @filename. And the power move: instead of writing the perfect prompt, let the AI help you write it.

A CLAUDE.md file is a document you write once, and the AI reads it at the start of every conversation. It's your AI employee's job description.

"Here's how my business works. Here are my brand rules. Here's what I NEVER want you to do. Here are the tools I use. Here's my writing style."

Teach it once, it knows every time. No more repeating yourself every conversation.

If you learn one thing, make it this. Plan mode means the AI writes out its full approach before making any changes. You review the plan, adjust it, approve it... THEN it builds.

It's like getting a proposal from a contractor before they start demolishing your kitchen. You see what they'll do, how they'll do it, and what it'll cost. Then you say "go."

90% planning, 10% building. Without plan mode, the AI rewrites half your files and you have no idea what happened. With plan mode, you're the boss reviewing a proposal.

Think of the AI's memory as a whiteboard. Everything you say, every file it reads, every response it gives — all goes on the whiteboard. The whiteboard is big, but it's not infinite. When it fills up, older stuff gets summarized to make room.

This is why the AI sometimes "forgets" what you said 20 minutes ago. The whiteboard filled up and your early notes got compressed. Long, rambling conversations fill the whiteboard fast. Short, focused sessions keep it clean.

Tokens are how AI measures text. Every word you send and every word the AI sends back costs tokens. There's a max per conversation (the whiteboard size). Type /usage to see your spending.

4 ways to keep costs down:

• Switch to a cheaper model for simple tasks (Haiku for quick stuff, Opus for complex work)
• Use /clear between unrelated tasks — don't let old context pile up
• Use /compact during long sessions — compresses the whiteboard without losing key details
• Use context management plugins to reduce token usage

AI gives you 3 brains to choose from:

• Opus — the smartest, most thorough, most expensive. Use for complex projects, big rewrites, anything where quality matters more than speed.
• Sonnet — fast, capable, affordable. Your daily driver for everyday tasks, quick edits, routine work.
• Haiku — cheap and quick. For simple questions, quick lookups, anything where speed matters more than depth.

You switch mid-conversation based on the task. Building a full marketing strategy? Opus. Renaming a file? Haiku.

Summarizes your conversation into a condensed version, freeing up whiteboard space without losing the key points. Like taking a photo of the whiteboard, erasing it, and writing a tight summary. You keep the important stuff, lose the fluff, and now have room for more work.

Wipes your current conversation. Use it when switching to a totally different task. Old context confuses new work. If you were working on email sequences and now want to brainstorm product names, /clear first.

Each conversation is a session. Close the terminal, come back tomorrow, pick up where you left off. Type claude --resume and you're back in your last session. Start fresh sessions for new topics. Resume old sessions for ongoing work.

You control how much freedom the AI has. From "ask me before every single edit" (safe, slow, good for learning) to "do whatever you need" (fast, you trust it, good for routine tasks). Press Shift+Tab to cycle through modes. When you're new, keep guardrails tight. As you build trust, loosen them.

Low effort for quick answers ("what time zone is Denver in?"). High effort for complex problems ("restructure my entire pricing strategy"). You don't always need full brain power. Change it anytime with /model.

Power move: type "ultrathink" in your prompt and the AI maxes out its reasoning for one response. It thinks through every edge case. Also, it's rainbow colored 🌈

Press Escape anytime to stop the AI mid-task. Like telling an employee "stop, new priority." It stops immediately and waits for your next instruction. No work is lost. You pivot.

VS Code is a free app that shows your files with nice formatting, tabs, and a sidebar. The AI runs right inside it. When the AI edits 5 files, you want to see what changed. VS Code shows you exactly which lines were added, removed, or modified — color-coded, side by side.

The terminal is great for giving instructions. VS Code is great for giving instructions and reviewing the results.

CLAUDE.md is your instruction manual. Your playbook. You write it. It's like handing a new employee a job description on day 1: "here's how we do things, here are the tools, here are the rules." It loads every conversation automatically.

Memory is the AI's personal notebook. It writes it (or you tell it to). It's like an employee jotting down notes after meetings: "boss prefers casual tone, the database field is called 'Video URL' not 'Drive URL'."

Hooks let you run custom scripts before or after the AI takes action. Think of them like triggers: when the AI saves a file, run a formatter. When it edits a Python file, run the linter. When it finishes a task, send you a notification.

They're defined in a settings file and run automatically — no manual intervention needed. This is how you turn the AI from "helpful assistant" into "automated employee that follows your company's processes."

Skills are like macros for AI. You define a workflow once (e.g., "deploy my website" or "write a blog post") and the AI follows that playbook every time. Skills combine instructions, context, and tools into repeatable processes.

Instead of explaining the same deployment steps every time, you create a skill. The AI reads it and executes. This is how you go from "telling the AI what to do" to "pressing a button and it's done."

MCP (Model Context Protocol) is how you connect the AI to external apps. Built-in tools work on your computer. MCP extends those tools to the internet: Airtable, Google Drive, Slack, Notion, and hundreds more.

Each MCP server is like a plugin. Install it, configure your credentials, and suddenly the AI can read your Google Sheets, post to Slack, or update your Airtable — all from one terminal window.

This is the cutting edge: instead of one AI, you run a team of specialized agents. One researches. One writes. One reviews. One deploys. They coordinate through a shared system and hand off work between them.

Tools like OpenClaw (what powers this site!) let you define agents with specific roles — a CEO agent that makes decisions, a research agent that finds opportunities, an execution agent that does the work. They run 24/7, check in with each other, and escalate to you when needed.

The ultimate power move: set up workflows that run on schedule without any human input. Cron jobs that trigger research. Webhooks that start processes. Agents that monitor channels and respond automatically.

This is where AI stops being a "tool you use" and becomes "a system that works for you." You set the rules, define the boundaries, and let it run. You only intervene for exceptions and approvals.

AI is powerful, but it makes mistakes. The #1 rule of working with AI: verify everything. Don't trust status claims without checking files. Don't trust "done" without seeing the output. Don't trust code without running it.

Build verification into your workflow: check file sizes, read outputs, run tests, compare before/after. The 30 seconds it takes to verify saves the 30 minutes it takes to fix a hallucinated result.