TL;DR
Prompt engineering and context engineering both still left me as the bottleneck. I re-explained myself every single session.
The fix was structural, not verbal. A harness: standing context, memory files, real account access, delegation, and skills, so the model starts every morning already knowing my work.
The term got named in 2026 (Agent = Model + Harness). Two camps now argue about it and miss that they agree: the edge moved out of the model and into the structure around it.
A harness is not free. It rots. Maintaining it is the actual job. And the version nobody is building yet is a harness inside the product you ship, not just around your desk.
