The Hofstadterian Codex — Where the Loop Began

Before the gateway, there was a protocol in a repo: how does a stateless AI agent stay itself across sessions, hand work to its future selves, and amend its own rules without breaking them? The Codex answered with files, intent-logged commands and structured handoffs — and everything I run now descends from it.

November 11, 2026 · Sven-Erik Nyberg

The Hofstadterian Codex: Teaching AI Agents to Think Across Time

The Problem Nobody Talks About If you’ve worked with AI coding assistants like Jules, GitHub Copilot, or Cursor, you’ve experienced this frustration: every new session feels like working with someone who has amnesia. You explain the project’s architecture, again. You re-establish coding conventions, again. You clarify the scientific goals, again. The agent might be brilliant in the moment, but it has no memory, no context, no understanding of the journey your project has taken. It’s like having a new genius contractor every day who’s never seen your codebase before. ...

September 5, 2025 · Sven-Erik Nyberg