The problem is that that idea gets misinterpreted as meaning that labor-saving devices can't reduce costs, when the idea (where it's true) is more about dead labor and surplus value being "baked into it" when capital is created. Tools facilitate labor, and they can transform the labor needed to do a task from intense, active participation to a role based more on guidance and pattern recognition.

For some commodities, reducing the per-unit production costs and increasing the volume a given amount of labor can produce will gradually devalue it and reduce the per-unit price (and in others it may increase demand as something suddenly becomes widely available and so consumption increases, where before its use-value was low despite its scarcity, because the whole idea of "supply and demand" is at best a vague suggestion instead of an actual fundamental principle), but for others the overall production cost and final sale price are entirely divorced because the per-unit production cost is negligible and the sale price is weirdly standardized.

So for digital media where individual copies are functionally infinite and nearly free but the creation of the initial work may have massive costs and the final sale price is going to more or less arbitrarily fall into one of a few standard brackets, then something like automated-but-curated asset creation isn't going to devalue the end product and may well be sold well below the cost of having what is functionally an artisan worker produce the assets instead. It would be sort of like using CAD and a milling machine to produce model injection-molds based on 3d models built from templates instead of having a skilled sculptor produce a physical master copy to cast resin molds from silicone or the like.