Youve Been Using AI the Hard W

The Shift from Browser to Terminal AI #

• The Inefficiency of Browsers: Using AI in a web browser leads to "context rot," where long chats lose their memory, and important information becomes scattered across dozens of tabs and notes.
• The Terminal Superpower: Running AI via a Command Line Interface (CLI) is faster, offers more transparency (e.g., showing exact context window limits), and allows the AI to interact directly with your local files.
• Privacy and Control: Browser-based AI creates "vendor lock-in." In the terminal, your project history is saved as simple text files on your hard drive, making your data portable and independent of the tool.

Google Gemini CLI #

• Accessibility: Gemini has a generous free tier and works across Mac, Windows (WSL), and Linux.
• Local File Interaction: Unlike the browser version, the Gemini CLI can read your directory, write research to markdown files, and execute Python or Bash scripts.
• Context Management: Using a /init command creates a gemini.md file that stores the project state. Every new session reads this file, allowing the AI to "pick up where it left off" without re-explaining the context.

Claude Code (Anthropic’s Terminal Tool) #

• The Power of Agents: Claude Code’s standout feature is the ability to delegate tasks to "sub-agents." These are fresh AI instances with their own context windows, preventing the main conversation from becoming bloated or biased.
• TUI (Terminal User Interface): Offers a richer visual experience within the terminal, including "Thinking Mode" and real-time web searching.
• Security First: By default, it asks for permission before reading or writing files, though this can be bypassed with a "dangerously" flag for power users.
• Output Styles: Users can create custom personas (e.g., "Brutal Critic" or "Home Lab Guru") that act as system prompts, tailoring the AI’s responses to specific niches.

Multi-Model Workflows and Syncing #

• Cross-AI Collaboration: You can run Gemini, Claude, and Codex (ChatGPT) in the same directory simultaneously.
• Unified Context: By syncing context files (like claude.md and gemini.md), you can have one AI write a draft, another critique it, and a third act as a research assistant—all working off the same local data.
• Automated Project Closing: Using custom scripts can summarize a day's work, update all context files, and automatically commit the changes to a GitHub repository for version control.

Open Source and Local Alternatives #

• OpenCode: An open-source alternative that allows you to use local models (like Llama 3) or log in with your Claude Pro subscription to avoid per-use API costs.
• Advanced Features: OpenCode supports session sharing (via URL), session "timelines" to jump back to previous states, and headless server attachment.

Security with Twingate #

• The Risk of Local AI: Giving AI access to your local machine is powerful but risky if remote access is required.
• Zero Trust Networking: Traditional VPNs are discouraged because they grant broad network access. Twingate is recommended for "Zero Trust" access, ensuring remote devices/users only access specific permitted resources.

Summary #

The video argues that moving AI workflows from the browser to the terminal is the ultimate productivity "hack." By using tools like Gemini CLI, Claude Code, and OpenCode, users gain direct file system access, the ability to deploy specialized agents, and total ownership of their project context. This setup allows for a more organized, "context-aware" environment where AI can perform complex multi-step tasks, critique work based on local frameworks, and maintain a permanent history through GitHub—all while avoiding the limitations and clutter of traditional web interfaces.