I read a great essay by Carson Gross today (link to the essay). In it, he argues that it is still important to write code, because it enables you to read code.

If you can’t read the code you are going to fall into The Sorcerer’s Apprentice Trap, creating systems you don’t understand and can’t control. — Carson Gross

In his opinion, you shouldn’t let AI generate code for you, but instead use it as a teaching assistant, except for small bits or code you actually don’t enjoy writing.

This is pretty close to the three rules I set for myself in my last post, although even more strict. From my own experience, I know that it’s difficult to implement this advice, because of what clients have come to expect in terms of development speed, but it’s still better to write some code manually and let AI generate other code, than to let it generate all of it.

---

I also read a post about how to read code by Sean Goedecke. It introduced me to the term dyadic scanning, which means doing several passes over a piece of text with different goals in mind, instead of reading it all sequentially.

This wouldn’t make sense for a book for example, but it does make sense for code, because it’s not written sequentially and it’s usually not read as a whole, but rather in terms of changes over time.

Dyadic scanning can be used to get a first overview, then figure out the sub-structure, and finally the details. After that, it makes sense to read the diff end-to-end.