Perhaps a more direct analogy would be that manual kerning is like hand-optimising code at a low level. You can improve almost anything to some degree, and an

Perhaps a more direct analogy would be that manual kerning is like hand-optimising code at a low level. You can improve almost anything to some degree, and an expert would be able to detect even quite small differences on close inspection. However, an isolated minor deviation would not make a significant difference in normal use, and the effort to hand-optimise everything is prohibitive. However, a user could notice either the cumulative effect of many suboptimal areas (poor built-in kerning in a font, careless inefficiency in frequently run code) or a single major problem (not tidying up a large logo or hero graphic, not optimising the body of a loop running many times causing a delay in UI response). In these cases, the benefit of manual refinement may be amplified enough to justify the effort.