Mark Meadows wrote an interesting piece at Robohub. Basically, on virtual assistants such as Apple's SIRI, Microsoft's Cortana or Facebook's M are "the testbeds for tomorrow’s personal robots":

"Our mobile devices are becoming natural language interface hubs for life management and, as a result, having a gravitational pull on an increasingly complex buzz of connected services and APIs. This means that things like search will change: we will no longer have to speak Googlese; paper and page metaphors will be supplanted by the more dynamic (and cognitively more addictive) character metaphor. And if trends in virtual assistants and intelligent helpers – software robots – continue, then knowledge-bases (such as Wolfram Alpha or IBM Watson) will continue to come peppered with a patina of natural language, allowing us to move through data faster, with less training, and in a more human manner.

[...] We can also foretell the future by looking at less advanced natural language systems. Bots – essentially natural language oriented scripts – are a good indicator of where the robotics industry is at because bots are pervasive, useful, and simple to author. TwitterBots and FacebookBots crawl through these systems like bees in a hive, industriously providing retweets, reposts, summaries, aggregations, starting fights and flocking to followers. They can be bought, auctioned, sold, and deleted; you can buy 30,000 Twitter followers on eBay for as little as for $20, provided they’re all bots."

Why do I blog this? Although I'm not sure whether these agents need a proper physical instantiation (bigger than a phone), Mark's argument is relevant; especially if you consider how talking to objects (interacting with voice, or chatting/tweeting to bots) becomes slightly more present (= less weird).