RefMOC: Andrej Karpathy

MOC: Andrej Karpathy

MOC: Andrej Karpathy

Map of Content for Andrej Karpathy's work, insights, and contributions to AI and coding agents.

Overview

Andrej Karpathy is a leading AI researcher and educator, known for making complex AI concepts accessible while building production systems at scale.

Background

  • Former Director of AI at Tesla: Led Autopilot development
  • Founding member of OpenAI: Early team member
  • Stanford: Created CS231n (Convolutional Neural Networks for Visual Recognition)
  • Educator: Known for clear explanations and educational tools

Current Focus

As of 2026, actively exploring and documenting the emergence of agentic AI systems and their impact on software development.

Content in This Repository

Threads & Insights

Projects & Tools

(None catalogued yet - add nanoGPT, micrograd, or other projects as we document them)

Key Themes

1. Agentic Engineering

Karpathy has identified "agentic engineering" as an emerging discipline - the practice of:

  • Setting up long-running orchestrator agents
  • Managing multiple parallel code instances
  • Ascending layers of abstraction
  • Providing high-level direction while agents handle implementation

2. Practical AI

Focus on what actually works in production:

  • Real-world examples (home video dashboard setup)
  • Honest assessment of limitations
  • Emphasis on verifiable functionality
  • Building intuition for task decomposition

3. Educational Clarity

Known for:

  • Making AI accessible to practitioners
  • Concrete examples over abstract theory
  • Open-source educational tools
  • Transparent discussion of both capabilities and constraints

External Resources

  • Twitter/X: @karpathy
  • GitHub: github.com/karpathy
  • Notable projects:
    • nanoGPT - Minimal GPT implementation
    • micrograd - Tiny autograd engine
    • CS231n - Stanford deep learning course

Related MOCs & People

Significance

Karpathy's voice matters in the AI coding agent space because he:

  1. Built production AI systems at Tesla scale
  2. Understands both theory and practice from research and deployment
  3. Communicates clearly about capabilities and limitations
  4. Documents in real-time the actual experience of using these tools

His February 2026 assessment of the December breakthrough is particularly notable as one of the clearest articulations of a widely-felt but hard-to-describe shift in software development practice.

This MOC will grow as we document more of Karpathy's work, projects, and insights.

Edit this page