Matt Pocock's Claude Code habits: 5 takeaways about the human, not the AI agent

Watched Matt on Nate Herk's podcast talking about his Claude Code setup. Most of it is about making agents better. The bits that stuck with me were about the human: Get grilled before you build. /grill-me only works if you actually answer. His own warning: saying "agreed, agreed, agreed" to forty questions and walking out with the agent's plan. I felt seen. Learn the brake before the accelerator. He runs four agents and 200+ commits a day, but he started in an empty folder, watching every file. He tried specs-only for months while the code got worse. If Matt Pocock needs months to call his own bad idea, I'm not catching mine on day two. Let the AI explain, then go read the source. His /teach skill warns that fluency gives "an illusory sense of mastery". A polite description of most of my year. Give it the work you hate. He hands off writing because he hates it. I'm keeping mine, because it's the thing I'm trying to get better at. Best line: "the clarity of their thinking ends up in the product they create." Rude, but fair. Video if you want it: youtube.com/watch Which of these do you actually do? I'm 2 for 5.