Uncle’s Diary · Learn AI Simply · 4 minutes · Three half-ideas, one bad report, and the question that sorted them out
I get harebrained ideas. Usually in the morning.
A half-idea used to be nobody’s business but mine. Now I have an AI that will take one seriously the minute I say it. That turns out to be a blessing and a problem.
Three ideas in four weeks
Early September. I was thinking about touring senior care homes in Asia, as what I am: a man who may want one some day. So we wrote down the questions I’d ask at the door.
Mid-September. I’d heard China is far ahead on care technology. Robots, sensors, updates that reach the family wherever they are. I said: let’s search for that every month and keep what we find. It got built and scheduled.
Early October. Which cities should we actually look at? I said: let’s score a few every week. That got built and scheduled too.
Each time it went the same way. I said the idea out loud, the AI built it, I said go, and I moved on to the next thing.
Three ideas. Three separate drawers. Nobody combing.
Boomer’s note: “The word is ‘harebrained’, after the hare, an animal of no fixed plan. Uncle spells it ‘hairbrained’. I have let it stand, because for once his mistake arrives with its own solution: a comb.”
Then one of them ran
On 7 October the city scout filed its first report. It was long. It scored three cities and filled in twenty-four boxes. Fourteen of the twenty-four said “Mixed.” Four said it couldn’t tell. There were about sixty links at the bottom.
And the three verdicts told me what I had already decided weeks before.
My entire review, word for word:
“This watch list scout that just ran didn’t seem to turn out too well.”
That isn’t an instruction. It’s a complaint. It turned out to be enough.
The comb
The AI didn’t defend the report. It listed what was wrong with it, and most of what was wrong was in its own instructions. It had scored cities I’d already ruled on. Half its questions can’t be answered from a desk. Its grading scale landed in the middle almost every time.
Then I asked the question I should have asked a month earlier:
“Let’s also understand ultimately what we want from all of this searching and databasing and stuff.”
It was three things, once I said them out loud. Share what we learn. Collect the good ideas we find that make life better for older people. And end up with a checklist anyone could carry into a care home.
With those three on the table, the three drawers turned out to be one drawer. The monthly technology search and the weekly city scout stopped doing each other’s work. The questions for the door became the checklist. Then the AI read what Medicare in America and Age UK tell people to look for, and found five things ours had missed: food, visitors, complaints, care plans and fire safety.
It still wasn’t right
The first checklist was about 55 lines. After the comparison it was about 100. I said:
“It’s too long, and it’s too long for a phone.”
Now it is 24 lines on top, with the rest folded underneath for anyone who wants it.
The lesson: you don’t need a good idea. You need a comb.
A harebrained idea isn’t a bad idea. It’s an uncombed one. Three things did the combing, and none of them was clever.
I said the half-idea out loud instead of waiting until it was finished.
I let it run once and looked at what came back.
When it came back bad, I asked why, and then asked what I wanted in the first place.
The third one is the whole thing. The bad report was the most useful thing that happened all month.
Try it
If you have a few half-ideas lying around, say this:
“I have three half-ideas that might be the same idea. Here they are. What am I really trying to get out of all of this? Ask me questions, one at a time, until we both know.”
The honest ledger — who did what
Me: three half-ideas, about a month apart. One complaint. One question about what it was all for. After that, three rulings: it’s too long, leave the statistics out, and make it public.
The AI: built all three jobs. Diagnosed its own bad report. Found and read the other checklists. Wrote ours three times.
The AI’s mistakes, left in on purpose: it scored cities I had already decided on. It let the checklist swell to about 100 lines. It told me the list was “about 85 lines” when it was about 100. And it wrote on the checklist page that this was “the list Uncle carries,” when I hadn’t carried it anywhere yet. It caught that last one itself.
Boomer’s verdict: “Three ideas, one month, no supervision. Then one poor report and one sensible question, and suddenly there is a checklist a stranger can use. I award 8/10. Two points off because the sensible question was available on day one, and I was available to ask it.”
The checklist itself is here: The Visit Checklist. Use it however you like.
New to all this? The Beginner’s AI Series starts from zero, and Lesson 6 teaches the checking habit behind that one complaint.
🐾 Join the Rescue Mission while he figures it out.
Figures counted 7 October 2026 from the scout’s first report and the three drafts of the checklist.