I think being an employee and launching a startup are on a continuum of sorts. With a startup you get lots of uncertainty, but lots of independence. There is a

I think being an employee and launching a startup are on a continuum of sorts. With a startup you get lots of uncertainty, but lots of independence. There is a job to do, you decide how to do it, and the market is your metric for success, for better or worse. As an employee you have a lot more security (worse case your perfectly steady paycheck disappears for a bit to be replaced by another, maybe smaller maybe larger steady paycheck) but most likely your performance is reviewed by a pointy-haired boss. As a graduate student, usually you get paid a pittance (sometimes you pay for the privilege), have even more uncertainty that a startup (after 5 years or more you may be kicked out; good luck transferring your credits) and your performance is reviewed by a committee of pointy-haired bosses. If you were to actually plug all that uncertainty into a utility maximization problem assuming a risk-averse individual, I seriously doubt it comes out as a good idea for any field, without maybe adding in a variable for "must prove myself more intelligent than the mundanes with a formal title bestowed by the intellectual aristocracy" Of course most don't really know that going in, and the article was right; once you are on that train it's hard to get off. All of this sounds a little bitter (I used to be a bright eyed idealist; I promise!) but I'm in the midst of slogging through my dissertation and things will be much better when (if) I can finally put on my ridiculous floor length hood. Anyone good with optimal control of vector-valued stochastic processes? Have I got a deal for you!