I agree with the general point in this article, but I think aiming the problem at the "API" versus a download button is slightly missing the culprit. I don't m

I agree with the general point in this article, but I think aiming the problem at the "API" versus a download button is slightly missing the culprit. I don't mind an API that allows me to easily request the entire dataset I want to grab with a single unauthenticated call. When this is possible, it's easy to create a download link for less-technical users by hard-coding a query into an href tag. The issue is when an API forces me to make a large number of queries (often tens or hundreds of thousands) to construct a relatively small dataset. Combined with rate-limiting and authentication, the whole thing becomes a nightmare to work with. Sure, the standard API interface is great if you want to programatically interact with an external database through a custom application. But that design does not lend itself well to accessing public data for the purpose of modelling and data analysis. But, it's not the API persay that's the issue. It's the way that APIs are designed.