Learning to code is so different in 2026, and in this video I break down what “coding” actually looks like now—because it’s not just you typing every line anymore. We’ve gone from “vibe coding” (throwaway weekend projects) to what Andrej Karpathy calls “agentic engineering,” where you’re orchestrating multiple LLM agents like a puppet master. But here’s the catch: to do that well, you still need real fundamentals. If you can’t read the code the AI generates, you can’t debug it, you can’t correct it, and you definitely can’t build anything reliable. So I lay out a practical learning path in roughly chronological order: (1) basics like variables, types, control flow, OOP, and APIs (I use Python as the default, but you can adapt it), (2) software architecture—tech stacks, system design, data flow, databases, testing, and deployment, (3) version control with Git + GitHub so you can track and revert changes (especially when agents go rogue), (4) security and privacy, which is a major blind spot for AI agents, and (5) a bonus topic: containerization/microservices so your environment stays sane and deployable. Finally, I share how to supercharge learning with AI: prime your resources with NotebookLM, use models (I default to Claude Sonnet for code) to explain snippets, generate analogies, and help you understand documentation (because I hate reading docs). And yes—projects are still the fastest way to actually learn.