I didn't reverse-engineer the protocol for my blood pressure monitor in 24 hours

Buried deep is this gem:

    I tried feeding a lot of this into various Als (Kagi gives you access to a few with a nice interface) and I found that they mostly were stupid in ways that made me think.

Perfect summary of why AI is useful for rubber ducking. Doing this with a co-worker you always get some questions that are a few layers too shallow for what you need (because they haven't been thinking about it more than a few minutes), but it always makes you think and really helps.

But the very next sentence is the important bit:

> A few times I thought they had "cracked the case" but actually they just made me waste time.

Despite trying various models and services for years now, I've come to the conclusion that the current iteration of AI services hurts more than it helps. I spend more time prompting, reading pages of babble, and trying to find signal in the noise that I would just chilling out and thinking through the problem.

The technology is neat but not productive.

Even when the duck doesn't answer at all it still often works.

I worked next to another engineer in the early eighties who was working on a different project. At tea break time we got into the habit of sitting opposite each other and explaining the problems we were having. Neither of us put much, if any, effort into trying to solve the other's problems because it often turned out that simply attempting to formulate the problem in a way that would make sense to the other party was enough to reveal the answer or at least a promising idea to try.