- I will write code until the day I die. It is an art form which I embrace, alongside many other art forms I enjoy creating in like electronic music, writing, essays, travel, photography, and much more.
It is a terrible mistake that some people have come to believe that code is only as useful as its final executable result, and not the art of expressing logic and meaning within a computer system purely on its own merits.
- You go! I don't think a lot of people believe that. No matter what, typing code will be around as a fantastic hobby and exercise. Just maybe not as a way to get paid.
- There are developer jobs that amount to feature factories and there are reliability jobs about getting things done. In this second camp, there is room.
- Something tells me the author doesn't seem to find joy when writing code, and how writing the code can make the architecture more sound.
- The author kind of gets there eventually:
> Chrome Extension Rewrite: Finally, I rewrote the element selector of the Chrome extension (the one that lets you select/pointer elements in the browser) the “traditional” way, by hand. Why? Because I realized that my struggles to explain it to AI were actually a problem of abstraction. Once I fixed that, the result was clean, solid, and perfectly ready for future iterations — this time with AI. I realized that getting these abstractions right was writing the spec. And that’s the kind of work architects do.
- In the footnote, the author linkedly suggests to "Read That F* Code".
Indeed, it is good to be able to read F* code.
So not sure why the author does suggest to be Typer and Thinker at same time. Thinking in Types(and categories) composes well with traditional logical thinking imho.
- Meh, sort of. Just because LLMs let you output reams of code, doesn’t mean you should use them to do that. As always, you should make the smallest diff that would accomplish your goal. Working this way, LLMs don’t really accelerate my workflow much except for work that’s truly boilerplate and for refactoring. But for the sort of small-ish changes that iterate towards product-market fit, I find I have to spend more time trying to get Claude to do what I want than just writing the code I need by hand.
- Just today it gave me a bunch of deprecated MongoDB calls and completely botched some async Python code. But it's definitely gonna be writing all the code soon. Just six more months...
- I’ve certainly found my bogus “5 hour limit” on the pro plan used up multiple times arguing with Claude about the simplest of concepts. So much in fact that I feel it’s by design to push users towards the Max plans… even if not true the fact that I think it at all is a loss for them.
- We already know software companies are intentionally making their products shittier to drive profits. Google making their search worse to increase the number of times people have to search (so they see more ads) is a good example.
There is absolutely no reason to not think AI companies aren't doing the same. Dial in the accuracy so that each tier is only so useful, constantly and subtlety encouraging you to pay a little more for just a few more queries because "the next prompt will make it work, I'm sure this time!"
