The CIO of a firm got on a call with me last week and described her situation with more precision than most leaders manage.

"I'm totally paralyzed by fear of not knowing what I'm doing, not knowing the things to anticipate, and just not knowing what the what is."

Twenty years running the organization. Managing hundreds of millions of dollars and a team of half a dozen investment professionals. And she was stuck at the very first move.

What broke it loose had nothing to do with a course. She watched a peer describe an AI that read her calendar, checked the travel time between two meeting locations, and pulled email background on the person she was about to meet. Her reaction: "I didn't even know something like that was possible."

That's the real starting problem. A sightline problem.

So here's the order I now walk people through in their first week. The order matters more than it looks.

Turn off "Help Improve our AI Models"

Before you do anything else, we want to keep our information private. By law, LLMs have to keep your data for 30 days, but then it gets deleted. If you have this toggled off. It's in Settings>Privacy under Preferences. This is Step 0 to getting started.

Write your context.

Before you ask Claude to do anything, tell it who you are. Your role, your business, how you make money, who your clients are, what you're working on this quarter, who reports to you, how you sound when you write. A page or two of plain prose. No formatting rules, no template.

This feels like homework. It is the entire game. A sharp person with bad context moves fast in the wrong direction, and so does a sharp model. Everything downstream gets better or worse based on that one file.

Put this in Settings>General>Instructions for Claude

Attach your connectors.

Calendar. Email. Drive. Whatever else your actual work lives inside. People skip this because it feels like plumbing, and it's the step that changes the character of the thing you're working with. Before connectors, you have a very smart stranger. After connectors, you have someone who has read your week.

Now, and only now, ask it what you should do.

Almost nobody makes this move. Once your context is written and your connectors are live, ask this:

Based on what you now know about me and my work, and everything you can see in my calendar and email, what are the three things I should hand to you first? Choose the ones where you'd save me the most time with the least risk of getting it wrong.

Then read the answer slowly, because it will beat the list you would have written yourself. You don't consciously track how many hours a week disappear into scheduling logistics, or into reformatting the same document for the fourth time. Claude, looking at your real calendar, does.

None of this takes a weekend. The context file is 30 minutes. The connectors are 10 minutes. The question takes thirty seconds and it's the one that tells you where to point.

The CIO I spoke with wanted to know what was possible for her, on a Tuesday, with the 13 people she has.

That's a question you can now ask directly.

Alex

PS Write the context file yourself the first time. Have Claude clean it up after. The thinking is the point, and it only takes an hour once.