5 Months Of Claude Code (As A Business Owner)

Key Concepts

• Claude Code: An AI-powered coding and automation tool used for business tasks.
• Claude Desktop App vs. VS Code: The desktop app is recommended for managing multiple project folders in parallel, while VS Code is better for advanced text editing and granular feedback.
• Claude.md: A configuration file that provides context to the AI; it should contain high-level workflow rules rather than just static data.
• Progressive Disclosure: A strategy of linking to external documentation files from Claude.md so the AI only reads relevant context when needed.
• Skills: Saved workflows (markdown + scripts) that ensure consistent output for repetitive tasks.
• MCP (Model Context Protocol): A standard for connecting AI to external tools (e.g., Notion, Google Sheets, Bento).
• CLI (Command Line Interface): A terminal-based tool that allows AI to interact with complex dashboards (like Google Cloud Console) more efficiently than manual clicking.
• Routines: Scheduled tasks that allow the AI to perform work automatically at specific times.

---

1. Claude Code Setup & Organization

The audit revealed that keeping all business tasks in a single folder leads to "context drift" and clutter.

• Granular Folder Structure: Instead of one massive folder, users should split projects into specific subfolders (e.g., Marketing, Support, Podcast).
• Desktop App Advantage: The Claude Desktop app allows users to open multiple folders simultaneously, each with its own Claude.md and specific context, preventing the "VS Code window overload."
• Interoperability: Users do not need to choose between the Desktop app and VS Code. One can start a task in the Desktop app for better organization and switch to VS Code via "Right-click > Open in VS Code" for complex editing.

2. Optimizing Claude.md and Documentation

The most significant takeaway is that Claude.md should be used for workflow rules rather than just storing static business facts.

• The "Self-Updating" Rule: Add a rule to Claude.md instructing the AI to proactively offer updates to documentation whenever it performs a task that deviates from existing guidelines.
• Mini-Wiki Approach: Create a documentation/ folder. Link to these files from Claude.md so the AI can reference specific guides (e.g., "Brand Voice," "Target Audience") only when relevant to the current task.
• Workflow Rules: Include "one-liners" to improve AI behavior, such as:
  • "If you can perform the task yourself, do not ask me to do it."
  • "Always verify and correct your own work before finishing."
  • "Infer the meaning of my prompts; don't take them too literally."

3. Advanced Automation & Tooling

• Moving from MCP to CLI: For complex tools like Google Cloud or Notion, use CLIs instead of MCPs. CLIs are more efficient for agents to navigate and provide cleaner data payloads.
• Scheduled Routines: Use the "Routines" feature in the Desktop app to batch tasks (e.g., processing member applications every morning).
• Human-in-the-loop: Configure routines to stop at the final approval step. This allows the AI to do 95% of the work (drafting emails, updating sheets), leaving the user to simply review and click "Send."

4. Notable Quotes

• "Instead of putting just context in your Claude.md, you kind of want to put rules. Rules that build context... you go one level of abstraction up and you just tell it how to maintain its own context." — Gael (Authority Hacker)
• "If you can do the thing, don't ask me to do it." — Recommended rule for Claude.md to reduce manual friction.

5. Synthesis & Conclusion

The core issue for many business owners is "mental friction"—the barrier to starting a task because of the setup required. By transitioning to a folder-based structure in the Claude Desktop app, implementing self-maintaining documentation via Claude.md rules, and utilizing scheduled routines, users can offload the "tedious" parts of their business to AI. The goal is to reach a state where the AI manages its own context and workflows, only surfacing to the user for final approval or high-level strategic decisions.