Andrej Karpathy’s 1-hour Stanford lecture on AI engineering might be one of the best explanations of how modern AI systems actually work. The progression he lays out is surprisingly simple: 10% → LLM 30% → Prompt 50% → Agent 70% → Loop 100% → Graph And that progression explains something important: AI engineering is NOT just about writing better prompts. A prompt is only one small part of the system. The real leverage comes from everything you build around the model: → Context → Memory → Tools → Data → Feedback → Execution loops → Multiple agents → Workflows and dependencies A basic LLM answers a question. An agent can take an action. A loop lets the system evaluate, improve, and try again. But a graph connects all of these pieces into a larger system where information, decisions, tools, and agents can flow together. That’s the shift happening in AI engineering. We’re moving from: “Give the model a prompt.” to: “Build a system that lets the model actually do the work.” One line from the lecture captures the idea perfectly: “Delete everything, keep Graph.” If you’re building AI agents, workflows, or AI-powered products, this lecture is definitely worth watching. Watch it once. Then bookmark it. You’ll probably come back to it. #AI #AIAgents #AIEngineering