Within the "what is the future of software engineering?" argument there is a "how much does one need to be a traditional software engineer to be productive with AI?" sub-argument. Opinions on that latter question seem to me much too polarized: there's a lot of "anyone can be a software engineer now!" and also a lot of "traditional software engineers are more necessary than ever." My view is more moderate: not everyone, but many more people, will be effective software engineers. There are simply too many people in (more or less) adjacent fields giving it a try, and it seems to me arrogant to assume that my expertise is inaccessible to all or almost all of them. My little phrase for thinking about this is "the gardener in Topeka;" I find it motivating to think of someone who currently is keeping a really awesome garden and happens to have the sort of intellect to be way better than me at software within a year or two.
Pangram says that my site is "100% human." I've run a bunch of my posts through it and haven't fooled it yet (everything here is, indeed, fully human).
Here is Ally Piechowski on auditing a codebase (via Simon Willison). It's very good. I'm struck by how much of this information can now be, or soon will be, available from AI. I'd guess that this will make the skill of conducting a good stakeholder interview more valuable (commodotize the complement!).
I'm slowly working my way through Andrej Karpathy's microgpt project and finding it rewarding. More generally, minimal working examples are a great category for learning from.