Link Details

Their software is way too expensive. (you don't see a price quote on any of their websites, there's a reason for that!). Also, it hearlds back to the old school methods of glitzy sell for big bucks deal. The cost per sale is way too high using that method. On the other hand, you get some nice eye candy for the bucks, the cash doesn't all go to the sales budget.

After reading this article it felt more like a sales pitch then something technical. I'm still not sure how palantir is akin to an OS except that (according to the sales pitch) databases are like block devices. IMHO operating systems do much more than allow access to data.

One of the big selling points of this sales pitch was that all data is mapped into the palantir ontology thus providing a common platform on which to develop machine learning algorithms. That's all well and good (my company does something similar, but in a more extensible way), but what happens when your customer gets new sources of data (intelligence) that don't fit in with that ontology? Do you have to force it into the system as something else? I remember an older intelligence system that only supported a limited number of types of intelligence. It was "extended" so that new types of intelligence were masqueraded as older types of intelligence. Consider an airplanes flight being represented as a cars route, some analytical methods just don't work due to relative speeds and physical barriers (like oceans and mountains). However, to the analyst both of these looked like car routes.

My point is this, any algorithm developed against a fixed ontology will eventually fail due to changes in the underlying sources of information or the information itself.