Prompt engineering was about phrasing. Context engineering is about what is in the window at all: which reference material, which prior steps, which tool results, and in what order.
The failure mode has changed with bigger windows. It used to be that things did not fit. Now they fit and the model attends to the wrong third, which looks like a reasoning failure and is actually a curation failure.
The practical discipline is subtractive. Give it the brief, the specification and the two most relevant examples, not the entire brand archive. Relevance beats volume at every window size we have worked with.
Is prompt engineering obsolete?
It is a smaller part of the job. Wording still matters; what is in the window matters more, and a well-worded request over the wrong material still fails.
Does a bigger context window remove the need for curation?
No. It changes the failure from "it did not fit" to "it fit and the model attended to the wrong part", which is harder to spot.