Youve Been Using AI Wrong—Here’s the Terminal-Powered Workflow That Changes Everything

TL;DR: 📹 Watch the Complete Video Tutorial 📺 Title: You’ve Been Using AI the Hard Way (Use This Instead) ⏱️ Duration:…

If you’re still using AI through your browser—typing prompts into ChatGPT, Claude, or Gemini web interfaces—you’re working slower, less efficiently, and with far less control than you could be. In this comprehensive guide, we’ll walk you through exactly how to harness AI directly in your terminal to supercharge productivity, manage complex projects, eliminate context loss, and even run multiple AI agents simultaneously.

Based on a detailed walkthrough from a seasoned AI power user, this article extracts every tip, tool, command, and insight shared in the original video transcript. You’ll learn how to install and configure terminal-based AI tools like Gemini CLI and Claude Code (Cloud Code), leverage project-aware context files, deploy autonomous AI agents, synchronize workflows across multiple models, and secure your setup—especially if you’re working remotely.

Why the Browser is Holding You Back

Most people interact with AI through web apps: opening ChatGPT, asking a question, scrolling endlessly, losing context, and juggling multiple chats across different platforms like Claude and Gemini to verify answers. This leads to:

• Scattered research across 20+ chat threads
• Lost context when sessions time out or hit token limits
• Frustrating copy-paste workflows into notes apps
• No persistent project memory between sessions

The result? A chaotic, inefficient workflow that feels like herding digital cats. But there’s a better way—one that puts you in control of your data, context, and output.

The Terminal: Your New AI Command Center

The terminal isn’t just for developers. When you move AI interactions into the command line, you unlock powerful capabilities that browser interfaces deliberately hide:

• Direct file system access: AI can read, write, and organize files on your machine
• Transparent context windows: See exactly how much context you’ve used (e.g., “99% left”)
• Persistent project memory via local context files
• Automation through scripts and headless execution

Best of all? You don’t need to be a terminal expert. As the speaker says: “It’s not really scary. The terminal’s a fun place.”

Getting Started with Gemini CLI

Installation (Works on Mac, Windows, Linux)

Google’s Gemini CLI offers a generous free tier and works across all major operating systems. Here’s how to install it:

# On Linux/WSL or macOS with npm:
npm install -g @google/gemini-cli

# On macOS with Homebrew:
brew install gemini-cli

Tip: If you encounter permission errors, prefix the command with sudo (e.g., sudo npm install -g @google/gemini-cli).

Setting Up Your First AI Project

Create a dedicated directory for your project to keep everything organized:

mkdir coffee-project
cd coffee-project
gemini

Upon first launch, Gemini CLI will prompt you to log in with your Google account (a free Gmail account works). Once authenticated, you’re ready to go.

AI That Writes Files for You—No Copy-Pasting Needed

Try this prompt in Gemini CLI:

“Research the top 10 reputable sources on making the best coffee. Compile the results into a document named Best Coffee Method.md and create a blog outline.”

Gemini will ask: “Do you want me to create a file for you?” Say yes—and watch it generate real files in your project folder.

This is transformative: AI can now interact directly with your file system. It can:

• Write Markdown documents
• Update your Obsidian vault (since notes are just files)
• Run Bash or Python scripts
• Organize research outputs automatically

The Secret Weapon: Project Context Files

When you run /tools in Gemini CLI, you’ll see all available capabilities. But the real game-changer is the gemini.md context file.

How to Generate a Context File

Type / and press Enter. Gemini will:

1. Analyze your project directory
2. Read existing files
3. Auto-generate a gemini.md file with project context

View it with:

cat gemini.md

Why This Changes Everything

Every time you launch Gemini in that directory, it loads gemini.md as context. You can now say:

“Write the intro for blog post one in the coffee series.”

…and Gemini will understand the full scope—without you re-explaining anything. Even better: as you work, Gemini can update gemini.md with summaries, decisions, and progress notes.

Start a new session days later? It picks up right where you left off. No more lost context. No more scattered chats.

Real-World Example: Building a Video Script with AI

The speaker uses this exact method to create videos. Their gemini.md file for this project includes:

• Project status
• Major creative decisions
• References to other key documents

When they ask, “Where are we at in the project?”—Gemini provides a full update based on the context file. This turns AI from a chatbot into a true project collaborator.

Introducing Claude Code (Cloud Code): The Power User’s Choice

While Gemini CLI is great, the speaker’s daily driver is Claude Code (also called Cloud Code)—a terminal version of Anthropic’s Claude. It’s not free, but if you already pay for Claude Pro (~$20/month), you can use it directly in the terminal with your existing subscription—no API keys needed.

Installing Claude Code

npm install -g @anthropic-ai/cli

Then launch it in any project directory:

claude

Log in with your Anthropic account and grant folder access when prompted.

Claude’s Killer Feature: AI Agents

Claude Code introduces agents—autonomous AI workers that handle subtasks with fresh context windows. This solves the “bloated conversation” problem once and for all.

Creating Your First Agent

1. Type /agents to open the agent menu
2. Select “Create new agent”
3. Name it (e.g., “Home Lab Research Expert”)
4. Choose scope: project-specific or global
5. Define its role: “You are a home lab expert…”
6. Grant tool access (web search, file I/O, etc.)
7. Select model (e.g., Claude Sonnet)

The agent is saved as a file in your project and can be summoned anytime with @agent-name.

How Agents Work Behind the Scenes

When you ask your main Claude instance to delegate a task:

• Claude spins up a new AI instance (the agent)
• The agent gets a fresh 200,000-token context window
• Your main conversation stays lean and focused
• The agent works independently but reports back

This is like having an AI employee walk into the office with a fresh cup of coffee—ready to tackle a specific job without cluttering your main workflow.

Running Multiple Agents Simultaneously

You can deploy several agents at once. For example:

“Launch our home lab agent to research Proxmox servers. At the same time, deploy a general agent to find the best pizza in Dallas, and another to find the best gaming GPU. Compile everything into one report.”

Claude will spawn three agents, each working in parallel with their own context and tools.

Monitoring and Managing Agents

• Press Ctrl+O to peek at what an agent is doing
• Use Ctrl+C (twice) to interrupt and regain control
• Resume sessions with claude -r

Bypassing Permission Prompts (Use with Caution)

By default, Claude asks for permission before writing files or searching the web. To skip these prompts (e.g., during bulk operations), use:

claude --dangerously-skip-permissions

Warning: This gives AI full autonomy. Only use in trusted environments.

Output Styles: Custom AI Personas for Every Task

Claude Code includes output styles—custom system prompts that define how AI responds. Think of them as AI “personas.”

Creating an Output Style

/output-style new

Then define your persona:

“You are a home lab expert designed to help me create the best home lab possible.”

Once created, you can switch styles on the fly or set them as defaults for a project.

Real-World Use Case: Scriptwriting

The speaker uses a “Script Writing” output style optimized for video creation. This ensures Claude adopts the right tone, structure, and focus—without manual prompting every time.

Advanced Terminal Features in Claude Code

Claude Code is packed with power-user features:

Using Multiple AIs Together: Gemini + Claude + Codex

The speaker doesn’t rely on just one AI. They run Gemini CLI, Claude Code, and Codex (ChatGPT’s terminal tool) simultaneously in the same project directory.

How It Works

1. Open all three tools in the same folder (e.g., coffee-project)
2. Ensure their context files are synchronized:
   • gemini.md
   • claude.md
   • agents.md (used by Codex)
3. Assign roles based on AI strengths:
   • Claude: Deep writing and reasoning
   • Gemini: Fast research and fact-checking
   • Codex: High-level analysis and critique

Real Example: Crafting Video Hooks

The speaker gives each AI a different task:

• To Claude: “Write a hook with an authority angle → Authority Hook.md”
• To Gemini: “Write a hook with a discovery angle → Discovery Hook.md”
• To Codex: “Review both hooks and recommend the best”

All three AIs work in parallel, read each other’s outputs, and collaborate—without a single copy-paste.

Owning Your Context: Escape Vendor Lock-In

Browser-based AI tools trap your work in their ecosystems. But in the terminal:

• All project data lives in your folder on your hard drive
• Context is stored in plain-text Markdown files
• You can copy the entire project folder anywhere
• If a better AI emerges tomorrow, you can plug it in immediately

As the speaker puts it: “I own my context. No one can stop me.”

Automating Project Wrap-Up with AI Agents

At the end of each work session, the speaker runs a special agent called “Script Session Closer.” This agent:

• Summarizes the day’s progress
• Updates context files
• Archives decisions and outputs
• Prepares the project for the next session

To run it:

@script-session-closer

These agents are just files in an agents/ directory—making them portable, version-controllable, and reusable across projects.

Security Warning: AI + Remote Access = Risk

Granting AI terminal access also means it can interact with your entire file system—and potentially your network. This becomes critical if you’re a remote worker.

“If someone has remote access to your network via a traditional VPN, their AI may have access to everything—servers, NAS, file shares.”

The Solution: Zero Trust with Tailscale Alternative

The speaker recommends Twingate (note: transcript says “Twin Gate,” but correct name is Twingate):

• Implements Zero Trust Network Access (ZTNA)
• Grants access only to specific resources (e.g., marketing team → NAS only)
• Blocks access to sensitive systems (e.g., “super secret coffee recipe server”)
• Enforces device compliance:
  • OS must be up to date
  • Hard drive encryption required
  • Screen lock and firewall enabled
  • Block older OS versions (e.g., Windows 10)
• Integrates with EDR tools like CrowdStrike, Intune, Jamf
• Free for up to 5 users

For home labbers and businesses alike, this prevents AI (or compromised devices) from becoming a network-wide threat.

Headless Mode: Running AI as a Command-Line Tool

You don’t always need an interactive terminal UI. Both Gemini and Claude support headless mode—running as one-off commands:

# Gemini headless example
gemini -p "Find the best pizza places in Dallas" > pizza-results.md

# Use in agent definitions
"You will use Gemini in headless mode: gemini -p '[query]'"

This allows you to build agents that call other AIs programmatically—creating AI-to-AI workflows.

Practical Daily Workflow: Syncing Context Across Tools

To keep Gemini, Claude, and Codex aligned:

1. Maintain identical content in gemini.md, claude.md, and agents.md
2. Update all three files whenever major decisions are made
3. Use a “session closer” agent to automate this sync
4. Store all files in version control (e.g., Git)

This ensures every AI understands the full project state, regardless of which tool you’re using at the moment.

Why This Approach is Future-Proof

Browser-based AI is designed for casual use. Terminal-based AI is built for serious work. By adopting this workflow, you:

• Eliminate context loss forever
• Automate file creation and organization
• Deploy specialized AI agents on demand
• Maintain full ownership of your intellectual work
• Stay agile as new AI models emerge

As the speaker declares: “Once you see AI in the terminal, you’re never going back to the browser.”

Getting Started: Your Action Plan

1. Install Gemini CLI (free) or Claude Code (requires Claude Pro)
2. Create a test project folder (e.g., ai-test)
3. Run your first prompt and allow file creation
4. Generate a context file with / (Gemini) or /knit (Claude)
5. Try creating an agent for a simple task
6. Secure your setup with Twingate if working remotely

Final Thoughts: Reclaim Control of Your AI Workflow

The future of AI productivity isn’t in flashy web apps—it’s in the terminal, where you own your data, control your context, and orchestrate multiple AIs like a conductor leading an orchestra. By moving beyond the browser, you transform AI from a chat companion into a true extension of your workflow.

So grab your terminal, install Gemini or Claude, and experience what it feels like to work at 10x speed—with all your project intelligence safely stored in a folder you control.

Remember: Your AI work shouldn’t live in a black box. It belongs on your hard drive—organized, portable, and ready for whatever comes next.

Youve Been Using AI Wrong—Here’s the Terminal-Powered Workflow That Changes Everything

Buy this item