20 Rules I Actually Run My Business On. Free Playbook, One Rule Per Page.

The rest of the guide

4. Don't type. Use voice.Typing makes you compress what you actually want into a short prompt.
5. Give instructions one at a time.Stack five instructions in one message and the model will quietly drop or blend a couple of them.
6. Don't describe errors. Screenshot them.Your paraphrase of an error message drops the one detail that actually matters, usually the line number or the exact wording.
7. Don't copy-paste responses. Connect your tools.Every copy-paste is a place where you become the integration layer between two apps, by hand, forever.
8. Skip the tool buffet. Learn Claude Code or Codex first.Trying five AI coding tools for an hour each teaches you five shallow interfaces and zero real workflows.
9. Expensive models plan. Cheap models execute.The expensive model is worth the cost for judgment: architecture, review, catching a bad plan before it gets built.
10. Big project? Plan mode first. Always.The failure mode on big projects isn't bad code, it's an AI that started building before it understood the whole shape of the problem.
11. Cut sessions at 75% context.A model running on a nearly-full context window starts forgetting your earlier instructions and re-deciding things you already settled.
12. Do a task twice? Build a skill.The second time you explain the same task, you're paying the re-explanation tax for the rest of your life.
13. Don't download every skill. Less is more.Every skill you install is something the model has to read and weigh before it can help you, whether you use it or not.
14. Send worker agents to read the long stuff. Keep your main context clean.A model's context window is the scarce resource in a long session, not its intelligence.
15. Never accept done. Make it show you the live link."It should work now" is a guess, not a result.
16. Correct it twice? Make it update its own instructions.If you're correcting the same thing a second time, the fix isn't a third correction, it's a permanent rule.
17. Every failure becomes a written lesson it loads next session.A mistake that isn't written down gets repeated the next time you're tired and moving fast.
18. Give every project a memory file. Sessions start pre-briefed.Every new session starting from zero context means you re-explain the same project history over and over.
19. What this project is[one paragraph]
20. Current state[what's built, what's live, what's in progress]
21. Decisions already made (don't relitigate)[list]
22. Open questions / next steps[list] Read this file first, every session, before doing anything else.
23. Run a second agent to audit the first. Never trust solo output.The agent that built something is the worst judge of whether it's actually right, it's motivated to say yes.
24. Schedule your agents. My inbox triages itself every hour.If a task is genuinely repetitive, running it by hand every time is the manual labor you built an AI to remove.
25. AI won't replace you. Someone running agents will.The risk was never the model itself, it's the person who figured out how to point five of them at their actual work while you were still typing one prompt at a time.