you should become a politician. the question was about the AI not sorting it out.
and if you don’t mind, show me how you assign a document to one of 1500 groups in a 6 or 7 level hierarchy in less than two seconds without AI. i reckon you cannot and your eye roll was just the typical nonsense we can expect from devonthink employees when they are called out on the shitty-ness of the app. you guys can’t seem to face an honest look at your app. fine. but don’t insult customers especially when you are too stupid to actually answer the question. thank you.
Civility in discourse is highly recommended. 
Databases can differ in design for many reasons. My database for financial information has more groups and is deeper in hierarchical structure than any of my other databases. I can file any document into it in two seconds or less, because the structure is designed for easily recognized (by me) characteristics and dates of the items filed into it. I don’t use Classify at all, because the groups in that database do not have patterns of contextual relationships among the documents they contain that would lead to fruitful use of the AI assistant. Manual filing is easy.
The database in which I spend most time for writing and research isn’t as highly structured, either in the number of groups or in hierarchical depth. It contains tens of thousands of documents, and is organized into topical groups. Many of those groups have no subgroups. Only rarely does a group hierarchy go down more than three levels. But almost all of those groups contain documents that are recognized by Classify as having characteristic patterns of contextual relationships that distinguish them from each other, so that I do find the AI assistant useful for filing new content, especially when it suggests two or more groups that I think would benefit from replication of the new item into them.
That database originated some 11 years ago, when I first started using DEVONthink. Originally, it had a much more complex structure, with deeper hierarchies. Over time, I simplified the structure rather than making it still more complex. Originally, I had tried to distinguish nuances in content that could be used for structural design. But I found that didn’t actually make the information content more useful to me, but only resulted in more work as new content was added.
I decided that my purpose in creating that database was not to classify documents in detail (although that could be a legitimate purpose of a database), but to use the information in the database for research projects, which often required using the same information in different ways, and that my initial efforts to give considerable granularity to the organization of the documents didn’t pay off for the next research project, and might even confuse me.
DEVONthink provides tools in addition to structural organization to help one identify documents that are useful for a particular purpose. Those include searches (and smart groups), See Also and tags. With those tools I can quickly refine a set of documents that are highly useful references for a research project, without having had to spend a lot of time trying to do so a priori, in a highly granular organizational structure.
That’s why my research database evolved from organizational complexity to a much simpler structure with shallower hierarchies, and in that database Classify filing suggestions are almost always acceptable for my purposes. (More and more often, after glancing at a new document, I’ll invoke Auto Classify.)
But my financial database evolved in the opposite direction, with much more complex and deeper hierarchical structure, and Classify suggestions would usually be of little or no use, were I to bother to try them.
I find it generally quick and easy to file new content into both databases.