From Context Engineering to AI Agent Harnesses

Discussion of the shift from model training to orchestrating LLMs and compute primitives, with practical context-engineering techniques for managing context rot and reducing token cost.