I feel a lot of this comes down to player expectations of what counts as a "repetitive" puzzle, and perhaps how players interpreted The Witness.

Counting purely "raw mechanics" introduced in new panels then yes, you could argue The Witness did have a lot of repetitive puzzles. Once a new puzzle grammar was introduced then essentially entire areas were dedicated to expanding the complexity of how you could interpret that set of rules.

And yet I don't think there were many puzzles that you could say were the exact combination of ideas from another puzzle.

A significant part of the design idea in The Witness was in not showing you, the player, any set of rules but instead having you infer them from simple and deliberately restricted examples of those rules in use, and then testing your inference against larger and more complex versions of that same shape of puzzle.

Those later versions of the same puzzle shape would sometimes force you to rethink what you inferred and refine your understanding of a rule or sometimes show you what felt like an emergent property of that set of rules. "I didn't know I could trace the inverse of a shape too", "I didn't know pairs cancelled out", "I didn't know the shape could be rotated", etc.

This approach almost necessitates having a lot of puzzles that start from the simplest possible example, and gradually increasing in scope and complexity over a set of panels, which can end up looking like "a lot of the same kind of puzzle".