January 2, 2026
Manual vs. mob mentality
I wrote the manual Karpathy said was missing for agentic AI
AI “manual” drops and the crowd explodes: hype, eye-rolls, and a date typo
TLDR: A free guide for building AI agents landed to answer Karpathy’s “no manual” complaint, and the crowd erupted. Builders swear it boosts productivity, while skeptics dismiss it as fluff and nitpick the tweet date—spotlighting a bigger fight over whether plain language can truly be used like code.
Nicola Sahar just dropped a free “first principles” manual for building agent-style AI helpers (think command-line robot assistants like Claude Code), answering Andrej Karpathy’s dramatic “there’s no manual” moment on X. The guide promises big-picture rules like morphing plain English into actions, keeping systems consistent, and measuring real autonomy. The vibe? Ambitious, with a dash of “finally, some structure.”
But the comments turned it into a full-on town hall. One builder claimed “10x productivity” using these patterns, cheering the idea that natural language can be bent into useful commands. Then came the skeptics: “English isn’t code,” scoffed a critic, calling the whole thing “hallucinated nonsense.” Another commenter went for the jugular on a tiny detail—the Karpathy tweet date—declaring it was 2025, not 2024 and basically telling readers “good luck” with the manual. Cue the meme-y pile-on: recursion jokes (“ah yes, AI always circles back to recursion”) and a lone “Thanks for sharing” that felt like a calm emoji in a flame war. For context, Karpathy’s original post was a rallying cry about programmers mastering new AI layers amid chaos—no manual, just vibes. Now there is one, and the internet is split between 10x dreams and eye-roll reality. Read Karpathy’s post here.
Key Points
- •Nicola Sahar released a free manual called “Morphic Programming” for building agentic AI systems with CLI agents like Claude Code.
- •The manual is inspired by an Andrej Karpathy tweet describing a new programmable abstraction layer in software development.
- •It outlines nine first principles: morphability, abstraction, recursion, internal consistency, reproducibility, morphic complexity, end-to-end autonomy, token efficiency, and mutation & exploration.
- •Additional content covers system design (repo structure, git, context engineering), practical tips, psychology, and example commands.
- •The manual is MIT-licensed, with future additions planned and updates available via a public GitHub repository and social contact channels.