Human-AI Collaboration for Deep Work

This thesis proposes a concise, mechanistic behavioural‑mode model for guiding large language models during deep‑work tasks. The framework outlines how to stabilise and direct LLM behaviour across distinct cognitive modes without anthropomorphism or assumptions about internal architecture. It provides a practical structure for achieving consistent, high‑signal human–AI collaboration in complex reasoning contexts.