A few months ago I noticed something that bugged me enough to eventually write a book about it.
AI coding tools are genuinely good now. Copilot, Claude, Cursor — you can describe a feature and watch working code appear in seconds. That's not hype, it's just true. But I kept running into the same pattern, in my own work and in conversations with other developers: there's a growing gap between being able to generate code and being able to tell whether that code is actually good.
Not broken — broken is easy. It throws an error, you notice, you fix it. I mean the quieter failure mode: code that runs, looks reasonable, and is subtly wrong in a way you only catch if you already understand what "right" looks like. An unindexed database column that's fine at 10 rows and falls over at 100,000. An authorization check that works for the happy path and quietly leaks data the moment a null slips through. A component with three pieces of state where one would do.
AI won't stop you from shipping any of that. It'll happily generate it, confidently, in about four seconds.
So when I sat down to write a frontend development book, I made a deliberate call that I think goes against how most "AI coding" content is currently being written: no AI at all until the fundamentals are solid. HTML, CSS, JavaScript, React, Next.js — built by hand, understood completely, mistakes made and debugged the slow way. Only once that foundation exists does the book bring in Copilot, Claude, and Cursor — and even then, the framing isn't "here's how to prompt your way to a finished app." It's "here's how to read what the AI gives you and decide, yourself, whether it's actually right."
