People reach for a bigger model when the real problem is that they are handing it the wrong information. Context engineering is the work of deciding what goes into the model's window, in what order, and in what shape, so the answer you want is the easy answer for it to give.
The prompt is only the visible part. Behind it sits everything you chose to include: the retrieved documents, the running summary of the conversation, the tool output from the last step, the few examples you picked. Every one of those is a decision, and most teams make those decisions by accident.
A useful way to think about it is to treat the context window as a small, expensive desk. You cannot fit your whole filing cabinet on it. You get to place a handful of pages, and the model reads what is in front of it. If you put the right page on top, the work is almost done. If you bury it under ten near-duplicates, the model has to dig, and sometimes it digs up the wrong thing.
So the job is not to write clever instructions. It is to curate. You decide what earns a spot on the desk, you order it so the important material is easy to reach, and you cut anything that is just taking up room. Do that well and a mid-sized model will often beat a larger one that was fed a mess.