Then they add this part to their design, and lay out the schematic and the printed circuit board according to best practices by the manufacturer. Once the first boards return they put measure the signal levels and waveforms in their circuit to verify they are with the specifications and that they match the ones on the eval board. Then they will 'corner' test the circuit (corners are low/hi temperature range, low and high voltage range) and verify it continues to work according to specification at all the 'corners' (if it does then you are generally ok with assuming it will work at all points "inside" those for corners.)
There are people who are either in a hurry or don't care who wire something up according to the application note, power it up once and call it 'good'. I've seen a number of cost reduced 'clone' equivalents that meet that description. @kens has done a number of blog posts that show this sort of mentality in detail.
I think this is the hardware equivalent of expecting programmers to write bug free code on the first try.
Most of the time when a programmer wants to implement something a little complex and a little outside their expertise, they use an external library. Likewise, EEs will often buy a chip produced by a third party manufacturer and use that to handle it. But the abstractions available to EEs are often a bit more leaky than those available to us programmers, since they are more constrained by the laws of physics. and as a result that third party chip is still harder to use than a software library.