To succeed at Valve, you have to be a self-starter. No one will tell you what to do -- if you ask, people may give you an opinion about what they think you should do, but no one hands you a list of your five most important tasks for the next review period.
Beyond being a self-starter, you have to come up with ways to judge yourself, or to gather feedback from customers, partners (developers/publishers/vendors), or other employees. I think the comp system works reasonably well, but the feedback system has never functioned well for all employees. When an employee is having trouble, the system (which is really just a group of peers, sometimes guided by HR or more senior colleagues) is more focused and effective at gathering and communicating specific feedback for people.
Some people are just driven nuts by the uncertainty of this kind of management and feedback system. Some people thrive and delight in the absence of semi-annual self-evaluation forms.
At Valve, it's hard to get a grand project started. It's easy to get a clever, valuable, smaller project started. Turning the latter into the former requires great communication skills or a clever plan of laying out small projects to take the company in the direction you think is smart (and which is confirmed by the outcomes of prior small projects). There are no patrons to make things happen for you. I disagree with a few other (typically former employee- ) reviews that suggest there is secret management structure that controls everything. There are certainly more senior people (who may or may not be more experienced by years of work) who can give great feedback on what is worth doing, or how to do something. But there is incredibly little control, and shadow management is just not true. I think that is often a projection by people who can't believe the uncertainty of radical freedom.