I spent the last month setting up a new community for the MacSparky audience. I’m not going to walk you through the build today. That’s a piece for another week. What I want to tell you about is where the actual decisions came from.

I worked through option after option. Names, structures, pricing, rooms, access tiers, onboarding flow. By the end I had a draft set labeled A through H. Yes, eight versions. I went in deep.

Claude was useful for a lot of that. Skills, drafts, the small gears that turn busywork into something I could finish in a sitting. I’d describe what I needed and the robot cranked the wrenches for me. That part really did help.

The big decisions, though, didn’t come from any of those eight drafts. They didn’t come from any large language model. If anything, the models would have led me away from the answer.

The big decisions came from phone calls.

A call with Mike about which platform actually fits the audience. A call with Nick on whether to launch one community or two, which sounds tactical until you realize how completely it reshapes the next year of work. A few conversations with my biggest customers about what they actually wanted, in their own words, before I built around what I assumed they wanted.

Each of those calls reset something I had been confidently typing into Claude for days.

There’s a reason for that. The people I called know me. They know the business. They know the audience. They had context the model doesn’t have and won’t ever have. The model is an amnesiac eager intern. Useful for donkey work. Hopeless at holding the picture of what I’m actually trying to do.

I’m not anti-AI here. I’ve spent years building Field Guides on this exact topic. The lesson from this experience is just that the model holds the shovel. You, in your guts, and your trusted human advisors hold the map.

So here’s a question for you. What decision are you about to make on Claude’s advice that would change if you picked up the phone and asked someone who knows you?