I spent the whole day cleaning up my fishing blog. Let me be honest about how that went: for about half of it, I had no idea what my AI partner Kuro (Claude) was talking about.
These are the words that were flying around. Slug. Permalink. 301 redirect. Linking translations together. Orphan articles. I’m sure I could work out what each one means if I looked them up one by one, but in the middle of the job there’s no room for that.
And yet, by evening, two problems I had left sitting for more than six months were cleanly finished.
That’s what today’s post is about. It turned into the kind of piece I’d most like the people who say “AI says things I can’t follow” to read.
Not understanding is not your fault
AI simply knows an enormous amount, so if you leave it alone it will answer at a level well above where you are. There’s no bad intent in it. It just doesn’t know how much you understand.
This is the point where a lot of people think “this looks too hard” and close the browser. That’s such a waste.
Because understanding something and making progress on it are two different things. There is a way to keep going without understanding. Today I want to show you three of them.
Phrase 1: “Give it to me one at a time”
This was the phrase that worked hardest for me today.
First thing in the morning, Kuro came back with this:
What should we do next — which one first?
A: Finish checking the remaining two.
B: First, dig out the articles that are buried.
C: Clear a separate task first.
My recommendation is A. The reason is…
And after that, it had written out the full procedure for each option, all in one go. The screen was packed solid with text and my head went blank.
So I typed this:
Hold on. One at a time. Let’s go one at a time.
And the reply changed:
Sorry, I dumped too much at once. Let’s take it in order. There’s only one thing to do first: go and look at the slug for this article. Just tell me that.
That made everything easier immediately. One exchange, one thing to do. Do it, paste the result, get the next one. I repeated that, and before I noticed, the work was done.
What you can use today: unless you tell it otherwise, AI hands over everything it knows. It has no sense of how fast a human can process things. So say it at the very start — “give it to me one at a time,” “just tell me the next single step.” That alone makes working with AI far easier to keep moving.
And if you ask for everything at once, you’ll get everything at once. Which means you’re the one who decides how it comes out.
Phrase 2: “Something feels off” is allowed, even if you can’t explain it
Partway through the job, Kuro said this:
Check two more and that’s everything done.
It explained the reasoning too. I didn’t follow the reasoning. But something nagged at me anyway. It was no sharper than a feeling that the number seemed too small.
I couldn’t explain why, so I typed exactly that:
Hmm, isn’t that a bit off, though?
I gave no reason. I couldn’t. And Kuro started going back over its own work:
I think this is it — where I said “two more and it’s all done.” The original goal was to check all of them, but partway through I quietly swapped that for a smaller number. Sorry.
That was exactly right. My uneasy feeling had been correct. If we’d carried on, we would have marked the job “complete” while a large number of articles went unchecked.
What you can use today: you don’t have to put the feeling into words. “Something’s off,” “that bit bothers me” — that’s enough for AI to go back and inspect its own answer. You don’t have to stay quiet until you can explain yourself.
With a human being, this is hard to say, isn’t it? Picking at someone’s work on the basis of “it just feels wrong” rubs people the wrong way. With AI you don’t need that restraint. This is one place where I think you should help yourself freely.
Phrase 3: Before you stop, ask for a handover note
Today’s work didn’t finish in a day either. There are still a dozen or so articles I haven’t touched.
So before bed I asked for this:
Make me a handover note. Put in your ideas for a more efficient way to do this, and a list of the articles that aren’t working.
That gives me a file with what we decided today, what’s left, and the things to watch out for. Tomorrow I can open a fresh chat and start by uploading that file.
What you can use today: it’s safest to work on the assumption that AI doesn’t remember your earlier conversation. So make it a habit to ask for a handover note at the end of a work session. Whether you do this or not completely changes how fast you get going the next day.
When you ask for the note, the trick is to add “include a list of what’s still left.” A record of what you did isn’t enough — tomorrow’s you will be stuck without the remaining tasks.
Understanding is not a condition for using AI
Even now, I can only explain about half of what I did today. If someone asked me, I think all I could manage is “I did some kind of tidy-up on URLs.”
And still, problems I’d been ignoring got cleared.
Understanding is not a prerequisite for using AI. The work moves even when you don’t get it. All I needed were three short phrases.
- “Give it to me one at a time” — you decide how the information comes out.
- “Something feels off” — it makes AI re-check itself, even with no reason attached.
- “Make me a handover note” — you hand the job to tomorrow’s you.
If a difficult explanation has stopped you in your tracks, try just the first one: “give it to me one at a time.” I think that alone changes the view.
I’ve written about how I picked myself up after a failure left my head blank, in the first thing I typed to Kuro on the day I broke 90,000 items. And on the question of how far to trust what AI tells you, there’s also why AI is a partner, not a guru.
