Who is Andrej Karpathy?

Co-founder of OpenAI, former head of AI at Tesla, and one of the best educators in the field (his courses have trained a generation). He popularized the term 'vibe coding.' In May 2026, he joined Anthropic, the home of Claude.

Why do these principles change everything?

They save tokens, time, and quality: responses that fit instead of wordy walls, projects that don't go off the rails. The 'surgical' principle wasn't theoretical — it was born from a deployment that broke everything because too many changes were pushed at once.

The 4 principles I apply with AI (Karpathy-style)

Q: What are the 4 principles for working well with AI?

1) Think before responding (state your assumptions, lift ambiguity). 2) Simplicity first (the minimum that solves it, nothing speculative). 3) Surgical modifications (only touch what was asked, don't refactor what works). 4) Objective-driven execution (define 'we'll know it's done when…' before acting). This is a personal synthesis, inspired by how Karpathy teaches.

1. Who is Karpathy (and why it matters)

Hard to find more credibility on the subject. Karpathy co-founded OpenAI, led AI at Tesla, and is above all one of the best educators in AI (his courses and videos have trained a generation). He also popularized the term "vibe coding." In May 2026, he joined Anthropic — when someone of that caliber chooses a house, you pay attention.

OpenAIco-founder
Teslahead of AI
OpenAIreturn
Eureka LabsAI education
AnthropicMay 2026

A career that summarizes the recent history of AI — and a move that says a lot about Anthropic.

2. The 4 principles (my synthesis, in his spirit)

Honesty first: these are not "Karpathy's official 4 rules." This is my synthesis, inspired by his way of teaching work with agents. The idea matters more than the label:

Principle 1 Think before responding State your assumptions, ask if something's ambiguous, rather than diving straight in.

Principle 2 Simplicity first The minimum that solves the problem, nothing speculative.

Principle 3 Surgical modifications Only touch what was asked; don't refactor what's working.

Principle 4 Objective-driven Define "we'll know it's done when…" before acting.

Four simple habits — it's their consistent application that makes the difference.

3. Why it changes everything

In practice, these principles save you tokens, time, and quality: responses that fit instead of wordy walls, projects that don't go off the rails. My principle #3 ("surgical") isn't theoretical: it was born from a deployment that broke everything because too many things were pushed at once. Since then, every change traces back to one specific request. One mistake, one rule.