we may have plateaued by now, but so far it's still a net gain - a new person comes in, learns things, often by reverse engineering, often improves the automated process, sometimes leaves etc.
But what we do need is a better way to measure automation gains and losses - basically any process that is automated shows up as a net loss in human productivity, which is simply not true.
And vice versa any automated processes probably don't account correctly for the total energy and resource use and the cost of maintenance (both software and hardware).
