Context architecture beats documentation dumps
Dumping the whole corpus into an AI agent makes it worse, not better. The fix is architectural: each task loads a curated slice, not everything you have. Here is the method, and the same move at three different layers: specs, sensor data…