Is prompt engineering considered real programming in this model?

Yes. Within the Karpathy framework, prompts are programs written in natural language that instruct the 'LLM Operating System' on how to process information and reason through tasks.

When should I use Software 3.0 (Prompts) over traditional code?

Use Software 3.0 when a task requires flexible reasoning, natural language understanding, or rapid iteration where sub-millisecond latency is not the primary concern.

Can I combine these paradigms in a single project?

Absolutely. Modern AI systems are typically hybrid, using Software 1.0 for orchestration and safety, Software 2.0 for perception, and Software 3.0 for high-level reasoning and user interaction.

What is the Andrej Karpathy software framework?

It is a mental model that classifies software into three eras: Software 1.0 (human-written code), Software 2.0 (trained neural network weights), and Software 3.0 (natural language prompts for LLMs).

Software Paradigms (Karpathy)

Name: Software Paradigms (Karpathy)
Author: jona

byjona

0•

学習とドキュメンテーション

Applies Andrej Karpathy’s framework for Software 1.0, 2.0, and 3.0 to guide architectural decisions in AI-native development.

This skill implements Andrej Karpathy's influential mental model for modern software engineering, categorizing development into three distinct paradigms: traditional code (1.0), neural network weights (2.0), and LLM prompting (3.0). It provides a structured approach for developers and architects to determine the best paradigm for any given task—balancing deterministic logic, pattern recognition, and natural language reasoning. By treating the LLM as a new type of operating system, this skill helps teams build robust hybrid systems and communicate the fundamental shifts in how software is created in the AI era.

主な機能

01Framework for Software 1.0/2.0/3.0 paradigm distinctions

020 GitHub stars

03Decision guide for selecting the right paradigm for specific tasks

04Stakeholder communication tools for explaining AI evolution

05LLM as Operating System mental model mapping (CPU, RAM, Storage)

06Hybrid architecture patterns for combining code, weights, and prompts

ユースケース

01Explaining the strategic shift toward 'programming in English' to stakeholders

02Architecting AI applications by deciding between traditional logic and LLM reasoning

03Onboarding developers to the 'Software 2.0' and 'Software 3.0' mindsets

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add jona/ycombinator-skills software-paradigms-karpathy

For use in Claude.ai and ChatGPT

Download Skill