Andrej Karpathy talks about 'Claws'

Source: Hacker News

Andrej Karpathy talks about “Claws”

21 February 2026 – Andrej Karpathy posted a mini‑essay on Twitter about buying a Mac Mini (the Apple Store person told him they’re “selling like hotcakes and everyone is confused”) to tinker with Claws:

I’m definitely a bit sus’d to run OpenClaw specifically […] But I do love the concept and I think that just like LLM agents were a new layer on top of LLMs, Claws are now a new layer on top of LLM agents, taking the orchestration, scheduling, context, tool calls and a kind of persistence to a next level.

Emerging implementations

A quick skim reveals several smaller “Claw” projects:

• NanoClaw – core engine ≈ 4 000 lines of code, fits into both a human’s and an AI agent’s mental model, runs everything in containers by default.
• nanobot
• zeroclaw
• ironclaw
• picoclaw (lol @ prefixes)

These examples suggest a growing ecosystem of lightweight, container‑based agents.

Terminology

Karpathy has a knack for coining useful terms (see his earlier posts on vibe coding and agentic engineering). He argues that “Claw” is becoming a term of art for the whole category of OpenClaw‑like systems:

AI agents that generally run on personal hardware, communicate via messaging protocols, and can both act on direct instructions and schedule tasks.

Outlook

While the exact setup is still evolving, Claws appear to be an exciting new layer of the AI stack, adding orchestration, scheduling, persistence, and richer tool‑calling capabilities on top of existing LLM agents.

🦞