Deciding what an agent sees at each step — the discipline that replaced prompt engineering once context windows got large enough to fill badly.
A large context window is not a reason to fill it. Models attend unevenly across long inputs, so burying the important instruction among fifty irrelevant files makes it likelier to be missed, not likelier to be used. Precision beats volume well before the window is full.
In practice it means retrieving instead of dumping, summarizing instead of pasting, spawning a subagent instead of reading in-line, and putting the thing that matters at the start or the end rather than the middle. It is the highest-leverage variable in agent quality and the least discussed.