It Feels Like Black Magic. Learn the Fundamentals Anyway.

It still feels like black magic

I was an early adopter of GPT-4, and it blew my mind that the thing even worked. It still does. I understand how it works, at least at a level — under the hood it’s a very good stochastic parrot. Code is words, and these models were trained on an enormous pile of words, especially the useful ones that make computers run the world. Bigger models know more things and pick up more nuance. You don’t need the deep internals to use it well; what you need to know is that the magic is real, it’s genuinely useful, and it is not slowing down. If you haven’t really used it yet, that’s the first thing to fix.

But it’s a tool, and it always will be

Here’s the part I want people to actually understand: AI is a tool. The unlock isn’t AI replacing you, it’s the human-plus-AI combo, and that combo is too good to ignore. The way I think about it is simple. A human walks around with something like an incomprehensibly huge model already — twenty-plus years of real-world training and an actual world model running in your head. AI, for now, is mostly trained on text. Pair the two and you get something better than either one alone. So don’t hand your thinking over to it. Aim it.

Which is exactly why fundamentals matter more now

When I was a junior developer at Prometheus Group writing code for paying clients, I had to learn the syntax and the concepts. Not a script kiddie pasting snippets, but actually understanding the fundamentals so my code was correct and performant. That hasn’t gone away with AI; it matters more. There’s a clear difference between typing “make me an app” and writing a real spec that says “use a hashmap here so this lookup is O(1).” The first gives you slop. The second requires you to know data structures, operating systems, and statistics, which were the hardest and most useful classes I took in my CS coursework. AI will happily generate something that looks like it works. Knowing the fundamentals is how you steer it, catch what’s wrong, and architect a system that actually holds up, and that is exactly what companies will keep hiring engineers to do.

What this looks like in my actual life

I’m a heavy user now. Claude and ChatGPT, and their coding harnesses Claude Code and Codex, which honestly weren’t even possible until a few months ago. I have an OpenClaw assistant managing my reminders, todos, and calendar; I can hand it a small skill file and it controls my Apple Reminders for me. Tell me that in 2017 and I would have laughed. This portfolio is another example: I used this class as a forcing function to write, refresh my real work, and produce an artifact I can build on later. The harness tooling moved from “neat demo” to “genuinely runs part of my life” in about a year.

What I’d tell someone new to AI

So if you’re newer to this, here’s my honest advice. Learn computer science and understand the fundamentals. Don’t stress about memorizing syntax, because that’s the part the tool is genuinely best at. Build things. Make a real artifact, whether it’s a document, a slide deck, or a working app, and show it to people. Be eager, and look for problems in your own life to fix, because that’s the fastest way to learn a tool this powerful. It is a life-changing tool, but its real power belongs to people who combine it with strong fundamentals.