[arx1422]

Another point - Bloomberg by an large doesn't fail. This isn't some consumer "build fast, break things, and iterate" model. I can remember one time in my two decades using Bloomberg that something broke. For half a day this year the IB chat system went down. The world of finance practically ground to a halt. But that was the wild exception. Compared to most platforms Bloomberg may be a bit archaic but it is rock solid and thats way more important when you have billions on the line in realtime.

[raldi]

What's amazing about this is that (at least as of the time when I worked there, 2003-2007) they've always broken just about every rule about coding style, DRY, unit tests, and so on ("We're not writing textbooks here") and yet (or because of this) were by a wide margin the highest-velocity shipper-of features I've ever known.

And this was all done while meeting the highest standards of performance and reliability. How'd they do it? I'd say two things: heavy investment in human testers, and an amazingly-rich "immune system": Deploy a bug in your new version of a Bloomberg function? Last week's code is still running alongside it, and in seconds you can route that one function over to the old instance while keeping all the other terminal functions on the new version. And an automated system might do it for you before you even get paged.

[zero_intp]

In the telecommunications industry, they are notoriously demanding customers.