October 3, 2026
AI JS
A few days ago, I wrote a message that included the sentence: “AI output is unreliable and generic”. In that message, I did a bad job setting the context of exactly what I was referring to.
Fellow list member and AI expert Allan Edmond rightly pushed back about the generic part with this post on his Substack.
You should read the whole thing, but I’ll summarize his point as something like:
“If you ask AI a generic question, you’re going to get a generic answer. But if you ask AI a really specific question and give it a lot of context, the answer will be a lot less generic.”
Yes! I totally agree. That’s been my experience too.
However...
The situation I was picturing when I said “generic” was a little different from Allan’s example, and it has been cropping up with increasing regularity in the consulting space.
I’ll give an example from my own experience:
I get people emailing to say that they have created their own “AI JS” by uploading all my books and articles into a Claude project that also includes all the relevant info about their business.
Then, when they have, say, a pricing question, they additionally upload the details of a potential project and ask their AI JS something like:
“What prices would JS recommend I put in this project proposal?”
And the answer it spits out will look like something I’d say. But it’s only based on the stuff I’ve shared publicly, and the stuff I share publicly is general. It has to be because I’m not talking to an individual when I write it. It’s my advice as calibrated for a target market, not any specific person (certainly not the AI JS user). Therefore, AI JS will give the user the same answer they would have gotten by searching through my books and articles.
It will not be customized to the user’s situation even though they have provided every detail about their business and the client project, because the exact advice I would give this individual does not exist in the source material that they uploaded.
In other words...
Including the business and project info in the context of the question will surely cause Claude to tailor the answer to the specifics, but that’s Claude talking, not me.
Here’s the thing...
This AI JS approach falsely gives the impression that the answer given is the answer the real me would have given.
It is not.
Why not?
Because I encounter edge cases and nuances all the time that cause me to make exceptions to my general rules of thumb.
Would the AI JS sometimes - maybe even most of the time - give the same answer to the user that I would have given?
Maybe.
The tricky part is, the AI JS user has no way to know which answers are actually the ones I would have given and which ones aren’t.
Yours,
—J