Who Is Andrej Karpathy?
The former Tesla AI director and OpenAI founding member now working on pre-training research at Anthropic.
Andrej Karpathy is an artificial intelligence (AI) researcher and educator who joined Anthropic in May 2026 to work on pre-training research.
At Layer3Labs, we build AI agents for small-business workflows, so we follow how frontier lab researchers influence practical software development.
Before joining Anthropic, Karpathy was a founding member of OpenAI, served as Director of AI at Tesla, and founded the education startup Eureka Labs.
Where Andrej Karpathy Is Now
Andrej Karpathy works on the pre-training research team at Anthropic under team lead Nick Joseph, having joined the organization in May 2026.
On May 19, 2026, TechCrunch reported that Karpathy joined Anthropic to start a team focused on using Claude to accelerate pre-training research. Business Today confirmed that he started at Anthropic that same week. In a public statement on X, Karpathy wrote: "I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D."
Bloomberg Law reported that his position at Anthropic PBC centers on research and development (R&D) to help train new artificial intelligence models. Fortune noted that he sits on a team led by Joseph.
- Current role: Pre-training researcher at Anthropic
- Organization: Anthropic PBC
- Known for: Leading Tesla Autopilot vision, being a founding member of OpenAI, creating CS231n, coining vibe coding, and authoring nanoGPT
- Education: Bachelor's degrees in Computer Science and Physics from the University of Toronto, master's degree from the University of British Columbia, PhD from Stanford University
- Recent news: Joined Anthropic in May 2026 to accelerate model pre-training research
Academic Background and Early Teaching
Karpathy completed his undergraduate education in 2009, earning bachelor's degrees in Computer Science and Physics from the University of Toronto. He completed a master's degree at the University of British Columbia in 2011. He then entered Stanford University, earning his PhD in 2015 under the supervision of Fei-Fei Li.
At Stanford, Karpathy authored and served as the primary instructor for CS231n, the university's first deep learning course. The course enrollment expanded rapidly, growing from 150 students in 2015 to 750 students in 2017, as recorded by Wikipedia. He taught deep learning for computer vision at Stanford until 2015, when he left to help start OpenAI, per TechCrunch.
Before his AI work made him widely known, Karpathy ran the speedcubing YouTube channel badmephisto, where a generation of competitive cubers learned to speedcube, according to Fortune. In 2020, MIT Technology Review named him to its Innovators Under 35 list.
OpenAI Founding Member and Early Research
Andrej Karpathy was one of the founding members of OpenAI in 2015, as he notes on his personal website, karpathy.ai.
From 2015 to 2017, Karpathy served as a research scientist at OpenAI. For more on OpenAI's leadership, see our guide on who is Sam Altman.
In June 2017, Karpathy concluded his initial OpenAI tenure to become Director of AI at Tesla.
Leadership at Tesla Autopilot
In June 2017, Karpathy became Director of Artificial Intelligence at Tesla, reporting to Elon Musk, as reported by Wikipedia. In this position, he led the computer vision team for Autopilot.
TechCrunch described him as Tesla's top AI executive and the leader of the Autopilot vision team.
Karpathy announced his departure from Tesla on July 13, 2022, following a four-month leave of absence. In a post on Twitter announcing the decision, he wrote: "It's been a great pleasure to help Tesla towards its goals over the last 5 years and a difficult decision to part ways." In that statement, he noted that during his five years at Tesla, Autopilot graduated from lane keeping to city streets.
Second OpenAI Period and Assistant Development
Karpathy announced his return to OpenAI on February 9, 2023. During this second period, his social media biography stated that he was working on "building kind of a J.A.R.V.I.S at @OpenAI", as reported by TechCrunch.
A year later, on February 13, 2024, OpenAI confirmed his departure. Karpathy commented on his exit in a post on X, stating: "First of all nothing 'happened' and it's not a result of any particular event, issue or drama", as reported by Business Today.
Following his departure, Karpathy stated that his immediate plan was to work on his personal projects. For other researchers who have moved between labs, see our index of AI people to watch.
Eureka Labs and AI-Native Education
On July 16, 2024, Karpathy announced the launch of Eureka Labs, an artificial intelligence education company. On its official site, the organization described its mission: "We are Eureka Labs and we are building a new kind of school that is AI native."
Eureka Labs outlined the structure by stating: "The teacher still designs the course materials, but they are supported, leveraged and scaled with an AI Teaching Assistant." The company's initial planned curriculum was LLM101n, an undergraduate-level course designed to guide students through training their own AI.
TechCrunch reported in July 2024 that Eureka Labs planned to host both digital and physical cohorts for LLM101n, and it found no public filings of any investments in the company at launch. Eureka Labs is on hold while he works at Anthropic; he says he plans to return to education. In May 2026, TechCrunch noted minimal public updates from the venture, citing Karpathy's remark upon moving to Anthropic: "I remain deeply passionate about education and plan to resume my work on it in time."
Open-Source Repositories and Video Instruction
Karpathy has developed several open-source educational repositories that demonstrate the mechanics of neural networks in minimal code. His project micrograd implements backpropagation through reverse-mode automatic differentiation over a dynamically built directed acyclic graph (DAG), packaging an autograd engine and PyTorch-like neural network library into two parts of roughly 100 and 50 lines of code.
His repository nanoGPT provided a minimal codebase for training and fine-tuning medium-sized GPT models, serving as a rewrite of minGPT. In November 2025, Karpathy marked nanoGPT as deprecated and directed developers toward nanochat. To remove high-level framework dependencies, he created llm.c, an open-source project that trains LLMs in pure C and CUDA without needing 245MB of PyTorch or 107MB of cPython, focusing on reproductions of GPT-2 and GPT-3 architectures.
Through his personal site karpathy.ai/zero-to-hero.html, he published Neural Networks: Zero to Hero, a free multi-part video course covering machine learning fundamentals from scratch in code. Lessons include building micrograd, language modeling with makemore, building a GPT model, and building the GPT tokenizer.
On October 13, 2025, Karpathy launched nanochat, an open-source model coded in roughly 8,000 lines of PyTorch, as reported by Analytics Vidhya. Released under an MIT license, nanochat functions as an experimental harness covering tokenization, pre-training, fine-tuning, evaluation, and inference on a single graphics processing unit (GPU) node, carrying the project tagline: "The best ChatGPT that $100 can buy."
Vibe Coding and Autoresearch Work of Andrej Karpathy
In November 2017, he published an essay titled "Software 2.0", arguing that neural networks are a new way of building software, as recounted by OfficeChai.
In February 2025, Karpathy coined the term "vibe coding" in a post on X, writing: "There's a new kind of coding I call 'vibe coding', where you fully give in to the vibes, embrace exponentials, and forget that the code even exists." Fortune summarized the concept as: "Describe what you want plainly and let the model do the work." Fortune quoted that post in March 2025; we examine the method in our guide on what is vibe coding and our practical breakdown of the best vibe coding tools.
In November 2025, Collins English Dictionary selected vibe coding as its Word of the Year for 2025. In March 2026, Karpathy said on the No Priors podcast that he had barely typed code since December 2025, per Fortune. See also our profile of Boris Cherny.
In March 2026, Karpathy published autoresearch, a repository where an AI coding agent conducts model training experiments autonomously. In his test run, the agent completed 700 experiments over two days and discovered 20 optimizations that reduced training time on a larger model by 11 percent, according to Fortune. On April 4, 2026, he released LLM Wiki, a GitHub gist describing a pattern for personal knowledge bases maintained by language models.
Career Timeline of Andrej Karpathy Across AI Labs
| Organization | Years | Role | Primary Focus |
|---|---|---|---|
| Stanford University | Until 2015 | PhD Student and Primary Instructor | Author and primary instructor of CS231n |
| OpenAI | 2015 to 2017 | Founding Member and Research Scientist | Founding member of the research lab |
| Tesla | 2017 to 2022 | Director of Artificial Intelligence | Led the Autopilot computer vision team |
| OpenAI | 2023 to 2024 | Returned to OpenAI | Building an AI assistant ("J.A.R.V.I.S" per his bio) |
| Eureka Labs | July 2024 to present (on hold) | Founder | AI-native school; LLM101n course planned |
| Anthropic | May 2026 to present | Pre-Training Researcher | Claude-driven pre-training acceleration under Nick Joseph |
For more on Anthropic's leadership, see our profile of Dario Amodei. For developers evaluating the model family he now helps train, see our guide to Claude for coding.
Reported Compensation and Net Worth Records
No verified net worth estimate for Andrej Karpathy has been published by reputable financial news organizations. Major outlets such as Forbes and Bloomberg have published no official calculation of his wealth as of October 2026.
Anthropic PBC has not publicly disclosed Karpathy's compensation.
Fortune reported in May 2026 that nearly 2 million people follow his AI writing on X.
Frequently Asked Questions
- Andrej Karpathy left OpenAI twice during his career. He first departed in 2017 to lead the Autopilot vision team at Tesla. He left OpenAI a second time on February 13, 2024, stating publicly on X: "First of all nothing 'happened' and it's not a result of any particular event, issue or drama." He said his immediate plan was to work on his personal projects.
- Yes, Andrej Karpathy announced his departure from Tesla on July 13, 2022, following a four-month leave of absence. In his public announcement, he described parting ways as a difficult decision after five years directing the Autopilot computer vision team as Autopilot graduated from lane keeping to city streets.
- Karpathy is widely known for his technical educational content and leadership roles across frontier artificial intelligence labs. He developed Stanford's first deep learning course (CS231n), led the Autopilot vision team at Tesla, was a founding member of OpenAI, created educational repositories such as nanoGPT and micrograd, and coined the term vibe coding in February 2025.
- At Anthropic, Karpathy works on the pre-training research team led by Nick Joseph. TechCrunch reported on May 19, 2026, that Karpathy joined to start a team focused on using Claude to accelerate pre-training research.
- Eureka Labs is on hold while Karpathy works at Anthropic. After founding the AI-native education company in July 2024 to develop courses such as LLM101n, Karpathy stated in May 2026 upon joining Anthropic that he remains passionate about education and plans to resume his educational work in the future.
- No official net worth figure for Andrej Karpathy has been published by reputable financial outlets such as Forbes or Bloomberg. Figures published on celebrity wealth websites lack documentation, and his compensation package at Anthropic has not been disclosed.
Review Your AI Agent Deployment Strategy
Speak with our implementation team about evaluating agent reliability, data security, and operational workflows before deploying autonomous tools.
Book a Consultation