AI researcher and engineer. Co-founder of OpenAI and former Director of AI at Tesla. Known for deep understanding of neural networks and for explanations that make complex AI ideas unusually accessible. He also coined Vibe Coding.

Ideas & Concepts

  • LLM Knowledge Base — his workflow for AI-assisted personal knowledge bases.
  • Vibe Coding — writing software through intention more than syntax.
  • Agent Proficiency — Karpathy’s thesis that managing autonomous agents will become one of the defining skills of the 21st century, more than programming or prompting alone.

Agent Proficiency as the Core Idea

Karpathy describes a shift similar to what happened in software engineering two decades ago. Back then the value moved away from “can you implement a B-tree?” toward “can you architect a system from 40 open-source libraries correctly?” Now the value shifts again: can you maintain structured knowledge bases, give agents clear instructions, and evaluate their outputs?

The key point is that this is not prompt engineering. It is curation over creation, closer to being an editor-in-chief than a programmer. The 21st-century power user decides what enters the system, what gets compiled, and which agent gets which job.

His second thesis is that your data needs to belong to you: Markdown files on your own machine, no hard vendor lock-in. If your “personal AI” lives only inside a cloud product, you do not really have personalization, you have a subscription.

Sources