I wrote the manual Karpathy said was missing for agentic AI | Hacker News Reader

I wrote the manual Karpathy said was missing for agentic AI

github.com

43 pointsnick283741 comments

I've been building with CLI AI agents (Claude Code specifically) for several months and noticed some powerful patterns emerging that have 10x’d my productivity.

Stuff like …

1. Morphability - natural language as executable, morphable code 2. Abstraction - encapsulating tasks into reusable commands 3. Recursion - stacking abstractions for leverage 4. Internal Consistency - the immune system of your AI system 5. Reproducibility - crash-resilient by design 6. Morphic Complexity - knowing when you've over-engineered 7. End-to-End Autonomy - what your system can do without human intervention 8. Token Efficiency - maximizing useful work per token 9. Mutation & Exploration - controlled self-improvement

Link: https://github.com/nicolasahar/morphic-programming

its free and i dont need anything from you except genuine feedback

also included system design patterns, psychological tips, and example commands :)

You have source code for this kind of system?

i'm working on it. but the fastest thing you can do is download the guide (markdown file) and have Claude Code review / incorporate it into its own Claude.MD

English - or better put: human language - is not the "new code". Since the inception of programming a person could ask another to write code.

This manual is hallucinated nonsense.

The only interesting part is how people uneducated in computers and mathematics always seem to fall into the topic of recursion with AI

you can continue to believe in the old paradigm, or accept reality for what it is.

someone who abstracts themselves up will be able to move and ship 100x faster than you in the next 12 months.

LOL, this is the list to keep in your head for this so called "manual". Best of luck of those who will work through this. BTW, Karpathy made that comment in 2025 not 2024.

  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

AI in 2026 is really all about morphability.

If you aren't using multiple agents, subagents, and autonomous MCP abstractions to construct a detailed morphological model of your codebase, you'll never appreciate the sublime bliss of man-machine union that the enlightened among us here have come to know.

> If you aren't using multiple agents, subagents, and autonomous MCP abstractions to construct a detailed morphological model of your codebase, you'll never appreciate the sublime bliss of man-machine union that the enlightened among us here have come to know.

Is this serious or satire?

Without a winking smiley or other blatant display of humour, it is utterly impossible to parody an AI hyper-enthusiast in such a way that someone won't mistake for the genuine article.

https://en.wikipedia.org/wiki/Poe's_law

Genuine Agent Zen is when your instructions .md contains but a single line, "Do!"

Everything else will be dated by Monday.

That is so January 1. Get with the program. Your approach is obsolete. You will fall behind in the global arms race. It's almost January 3, it's time for a new methodology!

Pro-tip: move to an earlier timezone so you can get the real edge on your competition.

> Inspired by Andrej Karpathy's tweet[0] on Dec 26, 2024

The tweet was in 2025, not 2024.

[0] https://x.com/karpathy/status/2004607146781278521

From the author: "A Few Disclaimers (1)

Yes, this manual was AI generated. However, the core ideas, first principles, and outline for this manual are all ..."

1. https://github.com/nicolasahar/morphic-programming/blob/main...

Okay, so the submitted title is a lie? "I wrote the manual..." Would you consider changing it to something more honest?

Oh come on.

Come on and what? We are dazzled by this cool new tech and so now precision in speech no longer matters?

Human language is the new code; precision in speech is outdated and irrelevant

You almost had me there, I'll admit, but then I looked at your (short, new) comment history for a Poe's Law check. A much-needed perspective around here! Keep it up, and good luck staying on the right side of the site guidelines -- your shtick is close to the edge, but very refreshing if done well.

It's not me who decides. I just pointed out that the irrelevant date is related to the AI generated nature of this text.

Ah. My bad, yelling at the messenger. But the actual "author", who might also be the submitter (nick = Nicola?), has some explaining to do. There's a lie in the submission title, and the same lie in the github readme intro.

Thanks for helping alert us all to the sloppiness and deceit. And thanks to all who flagged.

i genuinely wrote the thing myself. i wrote it in notion and had lots of spelling/grammar mistakes and no formatting, so i asked claude code to put it into a markdown file and polish the writing. im not going to sit here and do this myself bc this is not my full time job and im just trying to get my ideas out into the world

The person you are responded to isn't the author of the post.

Fair point, but neither is this lying "Nicola Sahar" character.

what am i lying about exactly?

>"Used AI"

>"Wrote this in a day"

>"So please forgive any imprecision or inaccuracies"

Um, no? You (TFA author) want people to read/review your slop that you banged together in a day and let the shit parts slide? If you want to namedrop some AI heavy hitter to boost your slop, at least have the decency to publish something you put real effort into.

You are not talking to the author. The comment was a quote from TFA, written (or, well, prompted) by someone else.

I know, that's why I'm quoting the author and not the commenter, and why I said "you (TFA author)"

i genuinely wrote this in a day. ive been in ai for 9 years, well before chatgpt came out. i used Claude Code to turn it from my notion draft (spelling mistakes, no formatting, etc) into a well-formatted markdown file. you don't need to believe me, move on with your life. the guide is free and is meant to genuinely help someone use AI in a better way

i wrote the manual on notion and asked it to put it in a markdown file and fix my spelling and grammar. if you read the disclaimer full, i specifically state i did not use it for brainstorming or adding net new ideas

thanks i fixed this!

How did you get a wrong Twitter link? And the updated note has two off-by-one errors.

https://github.com/nicolasahar/morphic-programming/commit/c3...

thanks i fixed both!

It would be nice if there were a domain specific language that could help with the internal consistency problem

i agree. it's still very early with AI programming in general, so this might evolve in the next few years

Pls consider donating this to the Linux foundation and making tons of announcements about it.

Tag them in tweets too

Thanks, will do that!

Thanks for sharing

No. Just no. You wrote a manual for using AI for software development is all, limited to a specific approach.

You did not write a manual for applying agentic AI more broadly and generally, which is what it is about. You completely missed the mark.

right at the top of Part 2: "The examples I give are going to be software engineering/coding specific, but they can be applied to any digital task."

you can genuinely use these principles for anything you want to do on your computer. I don't just use Claude Code for programming.

Reply on news.ycombinator.com