I built a tool to help me be a better husband

This week Jason writes about a small tool he built to help him plan date nights with his wife, and the two questions he now asks before he hands anything to a machine.

The story

I built a tool to help me be a better husband

A note from Jason.

I want to be good at taking my wife on dates, and I am not good at it. I think about it too late, forget to research or make the reservation, and resort to the same old options.

So one night I told Claude exactly that. I did not write a clever prompt. I described the problem the way I would describe it to a friend, and asked Claude to build a tool that would help me. Simple as that. Claude wrote a brief, which I set up to run automatically every week. And voilà! The whole thing took about ten minutes.

Now an email arrives every week. It lists a few restaurants for Thursday and Friday. It includes a concert or other cultural activity, a map and a reservation link. It also recommends which night based on weather and outdoor potential.

I use this example with leaders because it's simple and honest. It shows a three-part pattern: a person starts the work, AI does the middle, and a person finishes it.

At our first Base Camp session I put two questions on a slide. The first is "can we?" That is a capability question, and every month the answer is more often yes. The second is "should we?" which asks what the work does to us and to our people. If efficiency is the only test we apply, the answer to the second one will almost always be yes as well.

But is efficiency the only metric?

The three parts still help when the case is hard. Be honest about what you are bad at. Keep the direction and the decision in human hands. Let AI do the research and drafting in between.

More this week

People rate comfort written by AI higher, until they learn AI wrote it

A note from Jeff.

In four preregistered experiments, people rated messages of comfort written by AI as more compassionate than the ones written by trained crisis responders. That work is from Michael Inzlicht's lab at the University of Toronto Scarborough, published in January 2025.

A separate team at USC found something similar. AI-written replies made people feel more heard than human replies did. Tell the reader the reply came from AI and the gain disappears. How people reacted depended on who they thought wrote the words.

I will say plainly that this is not settled. These are a handful of studies and they do not all agree on how much the label costs you. But it matches what I have believed for a long time: AI can write kinder words than I can, but only a person can mean them.

A model can coach well. It can write well. It is more patient at eleven at night than I am. What it cannot do is decide to love you. Somebody has to choose that, and the choosing is the part that counts.

So when you decide what to hand over this week, ask whether doing it yourself is part of how you show you care. If it is, keep it, even on the weeks when the machine would do it better than you.

The practice

Name the friction, then write the brief

Another note from Jason.

Wherever you feel friction this week, stop and ask whether AI could do part of it. That is the whole trigger, and it is the one I keep having to teach. Most people go hunting for a use case first, and they never find one.

Jeff told me that "reduce friction" is too abstract to act on, and he was right. So we named the four kinds worth watching for.

Once you have one, most people open an assistant and ask it for the answer. That is the weak version, and it produces the generic result people then blame the tool for. Here is what I could have typed:

Here is what I actually typed. It admits the weakness and asks to be interviewed first.

The admission is what makes the brief good. A model that knows what you are bad at can cover for it. A model that only knows what you want gives you the average answer. Any assistant on a free plan will do this, and it takes about ten minutes.

Higher ground

That is the first question God asks a person in scripture. The first people had hidden, and God asked them directly and waited for them to answer. Jeff sees being answerable as what separates a person from a tool, and as the root of judgment, mercy and love.

AI will never have to answer for the work you hand it. You will. So give away the research and give away the draft, and keep the part where somebody has to answer for it.

The next step

Not everybody should hand AI the same work, because not everybody is strong at the same things. Our free AI Edge assessment takes about four minutes. It names your strongest of seven archetypes, then tells you what AI absorbs and what it amplifies for someone wired like you.

Find your AI Edge →

Then hit reply and tell us the friction you picked, or what surprised you in your result. We read every one.

Climbing with you,
Jeff and Jason