@artificialcorner: The first 6 lessons are for all Claude users: 1. Separate the work from the system. Content is for people to browse, the skills and tasks are code, and they need opposite things. 2. Name folders and jobs for what they hold. A folder called "claude" is a lie the moment you use a second tool. 3. If you'll edit it later, make it a variable. Values buried in a paragraph mean rewriting prose every time a time or folder changes. 4. If you have many scheduled tasks, build a calendar. Claude limits you in rolling 5 hour windows, so six jobs at 7am spend your morning while you sleep. 5. Give related jobs the same report shape. Fifteen jobs reporting fifteen ways is fifteen puzzles before coffee. 6. Tell Claude to delete what it removed. A note saying "we no longer use option 3" still puts option 3 in front of it. Mine reached for a deleted option twice in one month. The next 6 are for Claude Code users: 1. Every line in your CLAUDE.md gets billed forever. It's sent at the start of every conversation. One audit prompt cut mine by more than 50%. 2. A rule in CLAUDE.md is a suggestion, a hook is a guarantee. When the context fills up, Claude quietly stops following rules. A hook runs every time. 3. Learn just enough git. Four commands. It turns "undo what Claude just did to 30 files" into one exact step. 4. Delete half your MCP servers. Mine were 181.1k tokens across 298 tools, none in use. If a service has a CLI, use the CLI. 5. Your best model should judge, not type. Three Sonnet subagents draft in parallel, your strongest model reviews and never writes. 6. You are running one Claude. Run twelve. Split a job by file ownership and the agents work at the same time instead of queueing. #claude #ai #chatgpt #aitools

Frank Andrade

Open In TikTok:

Region: ES

Monday 28 September 2026 18:19:22 GMT