When you work alongside an AI, this happens from time to time. You ask it to make you a file, or to look something up, and then, right in the middle of the job, it just stops.
Today I ran into it over and over. So I want to write down what that “the AI stopped halfway” thing actually looks like, and how I got past it, together with the exchanges I had with my AI partner Kuro (Claude). If you are staring at a frozen screen thinking “did I break it?”, maybe this helps.
What actually happened: it stops in the middle
The symptom was always similar. I would ask Kuro to turn an article into a file, or to go and look something up. Kuro would start working. And then, one step short of finishing, it would go completely silent.
The file that was supposed to appear did not appear. The search results that were supposed to come back did not come back.
At first I got nervous. Is it broken? Is it my computer? But after it happened a few times, I started to understand that this is apparently just a thing that happens sometimes when you work with an AI.
What I did when it stopped
What I did was very simple. I told Kuro it had stopped. That was all.
Then Kuro would say “sorry, let me do that again”, and redo the same job. The interesting part is that even when the first attempt froze, the second or third attempt often went through with no trouble at all.
So the best response I found was not to panic, and just say “one more time”.
Me: It stopped again.
Kuro: Sorry, let me do that again. (→ this time it worked)
It is almost anticlimactic how simple that is. But it works.
I asked Kuro why it stops
It happened so many times that I ended up asking Kuro directly: why does this keep stopping?
The answer was an honest one.
I cannot fully identify one clear single cause myself.
And then Kuro listed a few things that came to mind.
One was the file name. When it tried to write out a file with a Japanese file name (full-width characters), it froze more easily. When I switched to an English name, it succeeded on the first try. After that I started creating files with English names from the beginning, and things got a lot more stable.
The other was trying to do too much at once. When we tried to split a job into two stages, first get ready, then do the real thing, it would sometimes stumble at stage one. It turned out to be more reliable to skip the extra preparation and just run the job I actually wanted.
Why it suddenly got solved: narrowing down the cause
Something struck me today. After all that freezing, at a certain moment it just cleared up. The reason was that Kuro and I had managed to narrow down the cause together.
“A Japanese file name freezes. An English one goes through.” Once I noticed that, I had a rough idea of the cause, so I could actually do something about it. Instead of blindly repeating “one more time, one more time”, I could aim: this is probably the cause, so let me change this. That is why it suddenly got solved.
This may be true of computer and app trouble in general. When you can turn “it does not work for some reason” into “this is probably the cause”, you move forward all at once. And being able to do that narrowing down while talking it over with an AI was genuinely reassuring.
What I took away from this
AI is a very smart partner. But it is not a magic tool that works perfectly, one hundred percent of the time. Sometimes it stops halfway. Sometimes it does not go well. Today I understood clearly that this is fairly normal.
What matters is not throwing your hands up and deciding it is useless. When it stops, say “one more time”. If that still does not work, ask “why is this happening?” together and narrow down the cause. That feeling of a human and an AI getting over the bumps in a three-legged race is, I think, the knack of working with AI.
I am not good with computers, and the only reason I have come this far is that every time something stopped, I did not panic, and instead thought “okay, so what do we do?” with Kuro. Days that do not go well are still good learning.
Summary
- AI sometimes stops in the middle of a task (this is fairly normal).
- When it stops, first just ask for it again. It often goes through on the second or third try.
- If it keeps happening, narrow down the cause (for example: change a file name from Japanese to English).
- Work is more stable when you do not get greedy, and simply run the job you actually want.
- When you can turn “it does not work for some reason” into “this is probably the cause”, it gets solved fast.
Next time too, I will keep an honest record of muddling through with an AI. The things that worked and the things that did not, all of it together, is what working with AI is.
If you want more on the narrowing-down part, I wrote about reading your data with AI to find the cause. And for how I think about the relationship itself, there is AI is a partner, not a guru.
