Gavriel Cohen discusses NanoClaw, his minimalist AI agent framework built to solve security flaws and dependency bloat in the age of agentic AI programming.

The New Stack is a publication covering trends and technologies in cloud-native development, DevOps, and software delivery. Developers can learn about containerization, Kubernetes, and cloud computing, as well as explore topics such as microservices architecture, serverless computing, and continuous integration/continuous delivery (CI/CD) pipelines.

The New Stack

Gavriel Cohen built NanoClaw, a minimalist AI agent framework, after discovering security and architectural flaws in OpenClaw (formerly Clawdbot). Key concerns included unvetted dependencies, lack of OS-level isolation between agents, and an AI-generated codebase of ~400,000 lines that no one could realistically audit. NanoClaw's response is radical minimalism: a few hundred lines of core code, each agent isolated in its own container, and a total codebase small enough to fit in a single LLM context window. Cohen also shares evolving coding principles for the AI era, including reconsidering DRY, relaxing strict file-length limits, and keeping Markdown skill files free of inline code to reduce token consumption and errors.

NanoClaw’s answer to OpenClaw is minimal code, maximum isolation