Did they get it wrong? I've always viewed IFTTT, "function as a service" offerings, many of the automation systems we use for building and testing and deploying software, and others as applications of those ideas. Maybe not as full-featured or ubiquitous as we imagined at the turn of the century, but certainly still in the vein of software agents.
I think they did get it wrong. The current role of what we might (Just about) call "agents" could have been described without the 90s terminologies and conceptions. But that period's agents were supposed to be something quite different from RPC (and really, even from AJAX when it eventually arrived). It turns out that we mostly just got a slightly fancier version of RPC that has allowed us to build some cool services, but the 90s part of it has mostly dropped away.
I was working at UWashington CS&E during that period, and the air was buzzing with the idea of "code objects" that would actually "go out on the internet and book your next trip for you". Sure, we ended up with Expedia which is pretty great, but it's not at all what was being imagined back then.
We have quite prevalent “code objects” too, tweets etc. we sent out on the internet to solve our problems. They do act as agents though, and are sometimes not successful. And you do not code for operations but for virality in an execution machine that is being optimised for this operation via specialisation of its components to be collaterally, obsessively specifically extremist.
We had "tweets" in the 1980s. They were called "usenet posts" and they were not restricted to 280 chars.
The core/key idea of agents in the 90s was that the agent would actually interact with services on your behalf, and reduce the level of interaction you were required to engage in without reducing the specificity of the interaction. That has not happened.
