Short term, it seems to eliminate workers. Shortly after those workers are eliminated, truth on the ground has changed, and now there’s an ADDITIONAL cost to updating the software, which is harder and more expensive to do without the workers that have been eliminated.
This allows for rapid development of functional integrations delivering real business capabilities — without the overhead of an SDLC. It helps wrangle shadow IT tools into a manageable system and generates lots of structural data on how the actual users of the system interact with it to guide investment in feature development. The RPA scripts with little / no value will naturally fall into disrepair, while the valuable ones will be maintained / improved.
I promise you, it’s way cheaper this way. Software teams stop getting inundated with one-off feature requests that take a ton of work to coordinate, scope, evaluate and test — even if they never get built. Business users get to eliminate time-consuming manual processes. Also keep in mind the first jobs to get replaced by RPA are the ones that have already been shipped offshore... An RPA team for a large enterprise is usually only 2 or 3 engineers and an army of BAs and PMs — but they’re replacing a bunch of individual projects that no longer need to be staffed.