From the "Tales of interest!" files:
A couple years ago, in our old office at the top of the Monadnock building, I'm working on code for a product we're doing while Vitaly is working on a pentest for a big Rails client, and, like, 5 other people in the office are working on 5 other things.
I'm procrastinating while noodling through some stupid Riak thing and so I hear Vitaly say something about how some request he sent to the app produced binary gibberish in the response.
It turns out I'd rather noodle around with Vitaly's app than figure out how to make Riak do whatever basic thing I want it to do, so I walk over and look at the response. Vitaly has already noticed that the "binary gibberish" is actually corrupted in the response headers; the response is corrupt.
We both agree out loud that this is unusual and bad. A couple other team members walk over. In the span of about 4 minutes we figure out that the bug is triggered by inputs that include NUL bytes, and responses appear to be corrupted at the point where the input is, in the headers, echoed in the output. And a few reloads show that what looks like gibberish is actually stale server memory.
This is that nginx bug that came out back when. It's approximately as bad as Heartbleed, except that it only affects nginx (and it has a more complicated trigger condition, albeit one that applies to virtually every web application). And because we were all together in the same room when Vitaly found it, we isolated, analyzed, and weaponized it in minutes, rather than in hours or days.
(Ironically, someone else had noticed the same bug on the same day, and [fair enough!] got the reporting credit.)
Stuff like that happened all the time at Matasano. The in-person requirement is one the company is unlikely to let go of --- not that I have any say in it anymore. ;)