What you don’t hear about AI is that it doesn’t know how human minds work, what our limits are, what we can do that they can’t. It has no memory. This was hard to believe, but you have to tell it to keep things in some kind of memory, usually a Markdown file, then tell it to read it from another file. These are things humans just do. I may have to say it two or three times, but you will remember it if you’re a normal human.

I know there are places you can leave an instruction to read it when it starts, but sometimes it doesn’t do it. It could never, in its current state, figure out how to make a product that people want to use. Nor would a human be able to easily read the code it generates unless you work hard at teaching it how that works, and if you have to do that, you might as well do it yourself.

It never looks for prior art, in some contexts — but in others, it’s encyclopedic about prior art. You might find a function named returnError in one place that takes a string argument, where most of the other instances take an object that contains a string. How do I, as a human, work in an environment like that?

The users who vibe code who think they know how to make code people will want to use are making the same mistake student programmers make, they come out of school only having done student projects, which have very limited objectives. The real world is far more demanding. The real world populated by humans that is. Just some random thoughts as I try to create usable code working with Claude.

Perhaps what’s needed is a Developing Better Developers program for pseudo-humans.