Issue #47

Am I Aiming AI or Just Pointing It?

Aim is what I’m calling how much effort you put into guiding the AI. Basic aiming is the floor and gets you competent work, with no particular point of view. Ask most models what they think of your plan and you get an enthusiastic yes. What you want is for it to poke holes in your theory, or hand you questions you can go validate or invalidate on your own.

It doesn’t think for you, but it can be a thought partner. You can bounce ideas off it, give it two options and ask it to model both with pros and cons, and it will occasionally surface a corner case you never considered. But the judgment stays yours, and that part doesn’t go away.

This is where the math gets interesting.

I have been saying all month that AI is a multiplier. That’s close, but it undersells it, as multiplication only explains the zero case. Put in nothing, get nothing back. AI is really an exponent based on how you aim it.

One raised to any power is still one, so all the leverage in the world changes nothing.

When you let it fly blind (aim < 1) it gets ugly. Half-aimed work running through ten rounds of AI gives you a diminishing value. Enormous volume, decaying value, and nobody catches it for a long time because it all looks fine. That is slop, and the value approaches zero.

However, if you spend the effort to aim it (aim > 1) then it compounds, and there are two levels to it. Aiming your own work is one. Getting your team aimed at the same thing is the other, and that is the only line on the chart that goes anywhere interesting.

Here is what that looked like for me.

We have run a family pick’em league for NFL season for years. It started on paper, became a text thread, and eventually my wife asked whether we could just build a small app for the family. A micro micro SaaS (nanoSaaS?). The kind of thing you know is buildable and never actually build, because there is no time.

A year ago I would have asked Claude for a PRD, shown it to her, and gone back and forth a little. Call it 20 percent me and 80 percent Claude. It would have sounded like it, too. It may have solved the problem; it would not have solved it well. She could always tell. I would hand her these AI generated artifacts and watch her get quietly tired of reading them.

This year I actually built it. Not because the tools got better, though they did. Because I spent the time up front understanding what we actually wanted, and because I have a year of practice iterating with these things instead of accepting the first answer. The PRD sounds like me now. It describes the thing our family wanted instead of a generic AI app. When I hand that to Claude and the MVP comes back strong, because what it was working from was aimed well.

Our kids are starting to head off to college. We still have them engaged in the pick’em. That is the whole return on the aiming.

Three Places Your Aim Shows Up

The same shift shows up at work, in three places.

Judgment. Mine got better, because I can run more scenarios, gather more input, and poke holes in my own theory before I commit to it. I go looking for the data that would invalidate it. But you can also see other people’s judgment in the work now. You can see the difference in the work people circulate. The best ones treat the AI output as a baseline and put judgment on top of it.

Conviction. The research I value most has the least detail at the top. Bottom line up front, detail behind it, sources cited. I am not going to read every report; nobody is. But when the effort is visible, when it is clearly more than “Claude summarized this,” it is much easier to believe the person thought through the problem.

Trust. Judgment and conviction are what build trust, and over time evidence accumulates that you can trust how you’re aiming it. I do not want your raw AI artifact, because I could have generated that too. I want what you think after you’ve spent the time reading the artifact. I want to trust your baseline knowledge, then your judgment, then your conviction to make a case for what we should do.

That‘s the part the tools cannot reach. Not because they are not good enough yet, but because it is not the kind of thing that gets amplified. It is the kind of thing that has to be there first.

Before you start up your next AI session, write one sentence describing what a good answer would look like. That’s aiming it. The amount of effort you put into your aim is how much leverage you’ll get from using AI.

Manager School goes out every week to working managers. If this was forwarded to you, subscribe at managerschool.com

See you next week!

-Frank