[davidw]

But if they're small, lean companies, is there room for people to just go off and do random things with little eye to profits?

An example would be someone like Simon Peyton-Jones at Microsoft:

http://research.microsoft.com/en-us/people/simonpj/

Much of what he does may eventually filter down to the rest of the company, but there's a real risk that some of it won't, or that it'll be picked up by someone outside of Microsoft, etc...

How does that sort of thing work out with a number of small companies networked together?

[pg]

I've seen some very small companies do things that would count as research projects in some places. E.g. Etherpad.

[jules]

While Etherpad is somewhat research-y, it is a product. The difference between Etherpad and what SPJ does is about as big as between a mathematician working on car collision simulation and one working on abstract algebra.

[wingo]

Dunno, Paul. While SPJ does have a lot of relevance to current computing, his work is not a product like Etherpad was. I haven't seen YC companies moving towards or staying with fringe languages -- not that I criticize that business decision.

Maybe the YC model will reach that "how will we be computing in 10 years" phase at some point. But (perhaps with good reason?) it does not seem to be there yet.

[davidw]

Etherpad came and went pretty quickly though, at least in its initial incarnation - correct? Simon has been at Microsoft for quite a while. You can certainly do interesting things in a brief period, but some stuff requires just requires more time/capital/materials.

[krunaldo]

Storm is another good example of small projects going big (perhaps not research wise, but building new stuff).

https://github.com/nathanmarz/storm

They did some other stuff to accommodate their needs.