We're All Confused About LLMs. That's Not Our Fault.
Around me, I keep seeing the same three things.
People using the wrong terms with full confidence. People convinced LLMs are either magic or a looming threat. And people quietly scared they're already falling behind — nodding along in conversations, hoping nobody asks them to explain what they just heard.
None of these people are bad engineers. Most of them are very good at their jobs.
The problem isn't that the ideas are hard. It's that the pace is fast. Every day brings a new acronym, a new tool, a new panic. The fundamentals exist — they're just buried under the noise. When you can't tell a fundamental from a fad, the only honest response is: where do I even start?
This month alone
Take MCP — the protocol everyone told you to learn last year. A few weeks ago, its own maintainers shipped the biggest overhaul since launch: they threw out the session-based design because it kept fighting with load balancers. The community's verdict? "We arrived at just send a POST request." The REST crowd had been waiting twenty years for that moment.
Companies raced to adopt that standard. People spent months learning it. And the rewrite revealed that the "new" thing was, mostly, a renamed version of ideas that have existed for decades. The fundamentals never changed — the plumbing did.
Or take the engineering ladder — prompt engineering, then context engineering, then loop engineering, and now graph engineering. A new name roughly every six months, each one arriving with the claim that the previous is obsolete. Dig in and you find the practice is always older than the label; the work just got wrapped in another layer and renamed. One recent write-up asked it plainly: are these different disciplines, or one idea we keep renaming?
If you understand the core idea, those names stop being homework and start being decoration.
What I did about it
So I built a place to start. A small guide — not a list of 100 terms, not a hype tour. One clear path. It starts with the only idea that actually matters: an LLM is just text in, text out. From there it walks forward step by step — why it forgets, how its memory works, how it gains hands, and how it eventually runs on its own. Each step connects to the last, so you're never left guessing how we got from point A to point B.
No math. No jargon you have to Google mid-sentence. Just the path, in order.
If you've ever nodded along while someone talked about LLMs and hoped nobody would ask you to explain it — this is for you. It takes about ten minutes, starts from zero, and by the end you'll know the shape of the whole thing.
You're not behind. You just haven't been shown the path yet.