Andrej Karpathy’s one-hour Stanford talk on AI engineering is the clearest breakdown I’ve come across of what’s really happening under the hood. It unfolds like this: 10% → LLM 30% → Prompt 50% → Agent 70% → Loop 100% → Graph The punchline: AI engineering isn’t the art of crafting prettier prompts. It’s system design around the model—context and memory, tool use, feedback cycles, and how information moves through the pipeline. “Burn it all down, keep the Graph.” If you’re building with AI agents, this is required viewing. Watch → Save