Expanding on my earlier remarks, Henry Farrell suggests that firms should be expected to underinvest in training workers, because the workers can leave:
However, it may not work out so great for the employer, who is unlikely to be able to capture all of the rents from this training, as Matt’s own career (from Prospect to Atlantic to Center for American Progress to Slate) demonstrates. The American Prospect did all of Matt’s future employers a considerable service, by helping him address the fundamental gaps in his skill set, but they haven’t received any direct compensation for it. Probably, they don’t care about this that much (my understanding is that the fellowship that Matt was on is designed in the hope and expectation that many of its recipients will go on to do wonderful things elsewhere), but other employers, such as employers of machinists, may not be quite so public spirited. They will perceive the risk that they’ll train someone up, and he’ll promptly defect to another employer, and hence be disinclined to do much training unless they absolutely have to.
Is that really true? The concern here is that the firm is being asked to provide something that's of value to employees and employers alike (skills) but that the employee gets to keep the skills without guarantee that the employer will get to keep the employee. So we should expect underinvestment in training of entry level workers absent some special arrangement like the TAP Writing Fellowship conceit. But it seems to me that to the extent that the training is transferrable the employee is gaining something of real value, and the employer now has the ability to reduce cash compensation accordingly. Employers need to choose between paying a premium for already-trained workers, or paying lower wages to less-trained workers but bearing training costs. But either approach is perfectly viable. You could imagine a dynamic in which some firms specialize in the training-centric approach and others specialize in the poaching-centric approach. Alternatively (and here my story links back up with his) the firms could offload the training function: "German employers in e.g. the mechanical engineering cluster in Baden-Wuerttemberg train machinists, but they do so through collective schemes, which spread the costs and the benefits among most of the relevant potential employers."
Interestingly, either way you end up with the value of clusters. The cluser is crucial to making the collective approach organizationally feasible but it's also crucial to making the specialization approach work. Here in DC, some publications are known as good places for young people to get a start and they're willing to work for sub-par wages to do so, while other publications are more focused on recruiting mid-career people. It all works because of the existence of a deep labor market and an industry cluster.