Or alternatively the interactive use case to iterate on a solution as part of R&D, which I believe is the main way people use their products.

The only use case I can see for this is either very complex computations for which either no other vendor exists, or requiring so much resources to run that their platform is the most convenient option.

But at that point you'd in practice rather use a more specialized (and much less expensive) solution. The only way to solve this problem would be to restrict your input to a fix format to make their output more predictable. This means that you'd have to re-implement non-trivial parts of their platform. Therefore, if you want to consume their service by leveraging the full smartness and cleverness of their platform, your consuming application needs to be equally smart and clever if you want to do anything more useful than displaying their raw output in an iframe. Their platform is so sophisticated that it produces output in a non-deterministic format depending on your search terms.

This is an interesting paradox, I sometime call it the Wolfram paradox, here is what I mean: I've always found Wolfram Alpha surprisingly unhelpful and impossible to integrate into an enterprise application in a meaningful way in practice.

Wolfram cloud basic plan