Most of the argument about AI and code is about how much of it gets written for you. I’m more interested in what happens when you ask it not to.
These are guides I’ve commissioned for projects I want to build myself. The brief is the same each time: explain the thing, don’t hand me the finished code. Tell me what to learn, what to build, and how I’ll know it works. Where an idiom is genuinely unguessable, show it, because memorising trivia isn’t the lesson. Where the design decision is the lesson, give me a function signature and a failing test and let me do the rest.
What comes out sits somewhere between a tutorial and a code review. It knows the language better than I do, it’s read the documentation I haven’t, and it’s good at pointing at the mistake I’m three steps away from making. What it doesn’t do is write the project, because then I’d have a working project and no idea how it works.
I’m putting them here for two reasons. They’re useful to me while I’m working through them, and they’re a decent answer to the question of whether this stuff is only good for churning out boilerplate.