Andrej Karpathy Reveals Shift to Directing AI Agents, No Longer Writing Code Since December 2025
The OpenAI co-founder now 'expresses intent' to AI, which built a 400k-word knowledge system and ran 700 experiments overnight.
Andrej Karpathy, a foundational figure in modern AI as an OpenAI co-founder and former Tesla Autopilot lead, has publicly detailed a profound shift in his workflow. Since December 2025, he has not written a single line of code. Instead, he now focuses entirely on 'expressing intent' to advanced AI systems, directing them as an orchestrator rather than a programmer. This move is a powerful signal from a top-tier practitioner about the maturity of agentic AI—systems that can autonomously plan and execute complex tasks.
Karpathy has operationalized this philosophy by building several personal AI agents. One is a unified home agent that has replaced six separate applications. Another is a comprehensive, LLM-powered personal knowledge system containing 400,000 words of his notes and data. Most strikingly, he created an autonomous research loop that designed and ran 700 distinct experiments overnight without human intervention. He cites a 'growing gap' between AI's rapidly expanding capabilities and users' understanding of how to leverage them, positioning himself at the frontier of this new human-AI collaboration model.
- Karpathy has not written any code since December 2025, focusing solely on high-level direction of AI agents.
- His autonomous AI research agent designed and executed 700 experiments in a single overnight session.
- He built a unified AI home agent that consolidated six separate apps and a 400,000-word LLM knowledge system.
Why It Matters
It signals a near-future where professionals orchestrate AI agents to execute work, fundamentally changing technical roles and productivity.