Things I would NEVER do as someone who has spent 400+ hours on Claude

I don’t start every conversation from scratch. I use Projects. Each one has its own instructions, uploaded files, and context. Claude already knows who I am and what I need before I type a word. I don’t let Claude agree with me unchecked. I run the LLM Council - five advisors with different lenses, blind peer review, one final recommendation. It’s the only way to actually pressure test a decision. I don’t draft lead replies manually. A scheduled task checks Gmail every hour, reads the enquiry, drafts a reply in my voice, and saves it to drafts. I review and send. 30 seconds per lead. I don’t write meeting recaps. Transcript in, structured summary out. Decisions, action items per person, follow-up email draft. Two minutes. Not twenty. I don’t give Claude generic instructions. “Write in a casual tone” is useless. I write specific rules: banned phrases, sentence length limits, real examples of writing I like. The specificity is the whole game. I don’t rely on Claude figuring me out passively. I tell it directly. “Remember I run an AI agency in Manchester. Remember I prefer British English.” That context loads into every conversation automatically. I don’t write weekly reports for clients. A Monday morning scheduled task pulls data from Sheets, Gmail, and Calendar — writes the report, saves it to Drive, drafts the email. It’s sitting in my inbox before I open my laptop. I don’t use Claude as a chatbot. I’ve connected it to Gmail, Calendar, and Drive. It reads my actual data and takes action inside those tools. I don’t install skills once and forget them. Skills are what turn Claude from a generalist into a specialist. Frontend design, document creation, slide decks I’ve built my own and I add to them constantly. I don’t treat Claude as a tool. I’ve set it up as infrastructure. Comment “Rules” and I’ll send over the Full Claude ‘Dos and Donts’ 🔥👇