I've always felt that pair programming is more useful on early stages of a task, where there is enough doubt about implementation details and discussing them is worth.

This way it felt more of a meeting between two persons discussing details first, while testing them live to check if we were on track second, instead of programming first and discussing second.

By the time we stand on the screen without talking too much we just stepped aside and separate the task if needed.

Any other kind of forced pair programming feels wrong, either because the task was already planned enough to no create enough discussion, or because it was small enough and the discussion was not worth. I've found myself on situations where "we needed" to make a task in pair programming and was dull as you say.