☀️ AI Morning Minute: Andrej Karpathy
He helped build the technology. Then he spent years explaining it to the rest of us.
Most AI researchers stay inside the lab. Andrej Karpathy spent years at the frontier of the field, then walked away from two of its biggest institutions to teach the internet how any of it actually works.
His YouTube videos on neural networks have millions of views. His GitHub repos are the starting point for a generation of AI engineers. And the term he coined in early 2025, vibe coding, became the name for an entirely new way of writing software.
Who they are
Karpathy is a Slovak-Canadian AI researcher who co-founded OpenAI in 2015, led Tesla’s Autopilot vision team from 2017 to 2022, and returned to OpenAI for a second stint from 2023 to 2024 focused on synthetic data and model training. He holds a PhD from Stanford, where he also co-created CS231n, the university’s first deep learning course.
In May 2026 he joined Anthropic as a researcher working on pre-training, focused on using Claude to accelerate Claude’s own training research.
Why they matter
Karpathy has a track record of being at the right place at the right inflection point. He was in the room when OpenAI was founded, led the team that made Tesla’s self-driving vision work at scale, and is now working on the part of AI development that most directly shapes how capable future models become. Pre-training is where the foundation gets laid.
His public work has genuinely changed who can participate in AI. The nanoGPT repository, a stripped-down implementation of a GPT model in a few hundred lines of code, is the clearest explanation of how language models work that exists. It’s been forked over 40,000 times on GitHub.
He coins terms that stick. Vibe coding started as a post on X in February 2025 describing a new way to program: tell the model what you want, let it write the code, stay in the loop but don’t touch the syntax. The entire ecosystem of AI coding tools now operates on that idea.
What they’ve said or done
When Karpathy announced his Anthropic move, he wrote:
“I think the next few years at the frontier of LLMs will be especially formative.”
He’ll work under Nick Joseph on pre-training research, with a specific focus on using Claude to improve Claude’s own training process. That work sits at the edge of what the field calls recursive self-improvement: AI systems that make themselves better.