Andrej Karpathy

Why we read him as an Architect

5 moments from his public record, and what each one shows of the Architect pattern.

• In a single essay published on 11 November 2017, titled "Software 2.0," Karpathy argued that neural networks were not just another tool but the start of a different way to write software: instead of a person specifying the logic by hand, a program is found by training on data, so that "the optimization can find much better code than what a human can write." He named the pattern before building anything to teach it. [2]
• At a CVPR workshop keynote in June 2021, Karpathy announced that Tesla, where he was director of AI, had already begun shipping cars with no radar at all. "We deleted the radar and are driving on vision alone in these cars," he said, arguing the camera-based system had become roughly a hundred times better than radar and that radar had started to "contribute noise" rather than help. [3]
• In 2020 he released micrograd, a back-propagation engine written in about 100 lines of plain Python, and in January 2023 he released nanoGPT, which trains a real GPT-2-class language model in roughly 300 lines of training code and 300 more for the model itself. Each repository is small enough that a single person can read the whole thing in one sitting and see exactly how the pattern works. [8]
• On 16 July 2024, Karpathy announced Eureka Labs, an "AI-native" school built around a "Teacher + AI" model, where a human teacher designs the course and an AI assistant personalises it for each student. Its first course, LLM101n, asked students to build a working chatbot-style language model from scratch in Python, C and CUDA. He founded a company to teach the pattern, not only to demonstrate it himself. [5]
• On 19 May 2026, Karpathy joined Anthropic's pretraining team, working under team lead Nick Joseph to use Claude to speed up the training of future models. "I remain deeply passionate about education and plan to resume my work on it in time," he said, closing a chapter running Eureka Labs to go back to frontier research inside an existing organisation. [6]

What made him so effective

Where his type’s weakness shows up

Every type has a known weakness. For the Architect it is this: can treat a finished design as finished work, when nothing has yet met a real user or a real load. Here is where that shows up in his record.

• The same June 2021 keynote where Karpathy announced Tesla's vision-only architecture came a few months before complaints of sudden, unexplained braking in Tesla vehicles climbed from 99 at the end of 2021 to 314 by February 2022, when the National Highway Traffic Safety Administration opened a formal investigation covering about 695,000 Model 3 and Model Y cars. [7]
• NHTSA's own account ties the timing of the complaints to Tesla's mid-2021 move from a radar-plus-vision system to vision-only sensing, the same change Karpathy had presented on stage as an improvement. [7]
• NHTSA closed the investigation on 2 July 2026, four years after opening it, citing Tesla's 2022 firmware updates, a sharp drop in complaints, and a finding of "low hazard to drivers and occupants." The design was not a failure in the end. It is still a clean example of the architect's blind spot: a pattern declared ready and shipped at a public keynote in 2021 produced a measurable, years-long federal investigation before real-world driving data showed it was safe enough to close. [7]

What to copy

Three things he is on record actually doing, that you can start doing this week.

1. 1

   Build the smallest complete version of an idea before you try to teach or scale it. micrograd proves the concept behind every modern neural network in about 100 lines a beginner can read in one sitting. [8]

2. 2

   Publish the essay that names the pattern before you build the company around it. Karpathy's "Software 2.0" is a single, freely available essay, not a product pitch, written years before Eureka Labs existed. [2]

3. 3

   State the reasoning behind a hard technical call in public, in enough detail that other engineers can check it, not just the headline conclusion. Karpathy's CVPR keynote explained the vision-only decision's actual logic, not only that Tesla had removed radar. [3]

What not to copy

Sources

Every numbered claim on this page links to one of these.

1. [1] Andrej Karpathy
2. [2] Software 2.0
3. [3] Tesla AI chief explains why self-driving cars don't need lidar
4. [4] karpathy/nanoGPT
5. [5] After Tesla and OpenAI, Andrej Karpathy's startup aims to apply AI assistants to education
6. [6] OpenAI co-founder Andrej Karpathy joins Anthropic's pre-training team
7. [7] Tesla Phantom Braking Investigation Closed by Feds
8. [8] Andrej Karpathy's Nano Repos: 120,000 Stars and One Teaching Philosophy