AI Engineering Roadmap
What It Wanted To Be
An exhaustive, ordered curriculum for learning AI engineering. Every concept mapped, every resource linked, every dependency traced. A complete path from “what’s a transformer” to “shipping production LLM systems.”
I spent weeks researching, categorizing, sequencing. I built Notion databases with progress trackers. I curated papers, courses, codebases. I wanted to know the whole map before taking the first step.
What Actually Happened
I planned more than I practiced. I optimized the learning path instead of learning. I collected resources instead of using them. The roadmap became a procrastination device — a way to feel productive without doing the hard work of actually building.
By the time I had the “perfect” curriculum, I had built nothing. And the field had moved on.
Why It Died
Theory trap. Planning felt like progress. Researching felt like learning. Organizing felt like mastery. None of it was.
The map is not the territory. I wanted certainty before action. I wanted to know I wouldn’t waste time on the “wrong” resource. But there’s no wrong resource — only not building.
Learning by building > building a learning plan. Ralph Gold taught me this by accident. I didn’t plan it. I just started. I learned by shipping broken code, not by reading about best practices.
The Lesson
-
Start before you’re ready. The first 20% of knowledge gets you 80% of capability. The rest comes from doing.
-
Plans are for revising, not following. A rigid roadmap is a liability in a fast-moving field. Build, then adjust.
-
The best curriculum is the one you finish. Perfect sequencing that you never execute is worse than messy execution.
The Irony
The “AI Engineering Roadmap” I never finished? A dozen better ones have been published since. By people who shipped first and iterated.
Planning is not a substitute for doing.
I thought planning would teach me. Doing would have been faster.