Mostly because automation isn’t making that person more valuable, it’s just removing it from that persons role. Skilled labor becomes unskilled labor. Supply goes way up. Clerks are a great example.
They used to do a lot for a store and had to know all the products. Now a clerk is just a last mile button presser. They don’t add much value at all. They’re not doing 4x the work. They’re doing 0.1x the work. Trying to say “but the value couldn’t be captured without that guy” is wishful thinking.
Not to detract from your overall point, which is absolutely correct, but having been a grocery clerk there's a slightly different phenomenon in play there. Having a clerk who knows the products well is a big help, but you really only need one available at a time. So rather than make acquiring that knowledge an actual part of the job, they rely on scale to have enough workers who learn it through their own initiative.
Which is bad for skilled labor (wrt. that set of skills) but good for unskilled labor.
The effect of productivity improvements on the real incomes of workers ultimately depend on how the change affects productivity of the marginal worker. You need to look at more than a single firm to figure this out. If the availability of spreadsheets makes clerical workers so much more productive that overall demand for their services increases more than four-fold (which might be definitely possible), the marginal spreadsheet-tweaking activity will be paid quite a lot, and workers' income will increase.