The foundation: context engineering and reliability

The starting point is reliability. When an agent must run for a long time and stay coherent, you have to contain the risk of compounding errors. If you are not careful, small mistakes accumulate and the system falls apart. At the core of reliability, Yan argues, is context engineering.

His framing of the term is memorable: prompt engineering was about writing your task in the ideal format for a chatbot; context engineering is the next level, doing this automatically in a dynamic system. He calls it "effectively the number one job of engineers building AI agents." This is the same conclusion the context-engineering articles reach, arrived at from a different direction.