我写了 Karpathy 说的 agentic AI 缺乏的手册。

我写了 Karpathy 说的 agentic AI 缺乏的手册。
I wrote the manual Karpathy said was missing for agentic AI

原始链接: https://github.com/nicolasahar/morphic-programming

受到近期人工智能工具快速发展带来的压迫感启发，Nicola Sahar 创建了一份免费手册，介绍如何使用 Claude Code 等 CLI 代理构建高效的 AI 系统。该手册解决了不利用这些新工具力量的“技能问题”——本质上，是一种强大但未记录的“外星工具”，正在重塑编程。它概述了“形态编程”的**九个第一性原理**，侧重于将自然语言视为代码，并构建适应性强、自主的系统。关键概念包括**形态性、抽象、递归和令牌效率**，以及有关系统设计、上下文工程，甚至与这些随机 AI 实体合作的心理学的实用建议。该手册旨在为开发者提供对这个新领域的根本理解，提供示例命令并鼓励社区反馈以供未来开发。它在 MIT 许可下免费提供。

## 代理AI“手册”在Hacker News分享 Hacker News的一位用户（nick2837）分享了一份使用AI代理（特别是Claude Code）构建的“手册”，灵感来自Andrej Karpathy的一条评论。该手册可在GitHub上找到([https://github.com/nicolasahar/morphic-programming](https://github.com/nicolasahar/morphic-programming))，概述了九种有效的代理AI开发关键模式：可塑性、抽象、递归、内部一致性、可重复性、形态复杂性、端到端自主性、token效率以及变异与探索。该帖子引发了争论。一些评论者认为这些模式很有见地，而另一些人则认为该手册是“幻觉产生的胡言乱语”或过于乐观的炒作。一个主要的争议点是作者声称*编写*了该手册，但后来发现它主要由AI生成。进一步的讨论集中在该领域快速发展的本质上，一些人认为该手册很快就会过时。尽管存在批评，但该资源仍然引发了大量关于在软件开发中利用AI代理的最佳实践的讨论。

原文

By Nicola Sahar

A free manual on first principles for building useful AI systems with CLI agents like Claude Code.

Inspired by Andrej Karpathy's tweet on Dec 26, 2024:

"I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There's a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind."

So I wrote the manual.

morphic_programming_manual_v1.md

9 First Principles:

Morphability - natural language as morphable code
Abstraction - tasks become reusable commands
Recursion - stack abstractions for leverage
Internal Consistency - prevent system drift
Reproducibility - crash-resilient design
Morphic Complexity - recognize over-engineering
E2E Autonomy - measure actual capabilities
Token Efficiency - maximize work per token
Mutation & Exploration - controlled self-improvement

Plus:

System design (repo structure, git, context engineering)
Practical tips and psychology
Example commands you can use

I founded and sold Semantic Health, an AI healthcare startup. Now exploring AI for consciousness and mental health.

More principles and advanced topics
A compressed version of the first principles you can add as system context in your own morphic system

Star or watch the repo for updates.

Open an issue or DM me. What did I miss? What should Part 2 cover?

MIT License - free to use, share, and adapt.

我写了 Karpathy 说的 agentic AI 缺乏的手册。 I wrote the manual Karpathy said was missing for agentic AI

我写了 Karpathy 说的 agentic AI 缺乏的手册。
I wrote the manual Karpathy said was missing for agentic AI