With modern messaging systems like Twitter, the simple re-Tweet function represents a sign of digital empathy. It's not something that a user is saying they agree with or disagree with necessarily: it's just a virtual nudge. Equally, people are sharing photographs of what they're doing online with such frequency that we can now recall images of places we've never even visited - and this in turn confuses our own history of where we've been, what we've seen and the way we recall memories. If we look at Ridley Scott's film Blade Runner, the 'replicants' that populate it all possess two lead qualities:

They are robots built upon other people's memories and they are in pursuit of empathy

So using this as a starting point, we developed a simple bot system that fed off Tweets and then used its own re-Tweeted messages as a source for uncovering new media and new data. These first bots were essentially recycling the internet and it was fun watching them do their own thing. We knew roughly what each one should be doing, but had no idea that it was going to pick a specific piece of music or a particular image. But then things got weirder. It turns out that if you leave a bot like that alone for a few days, an actual personality begins to emerge.

All our bots operate within boundaries of interest, much like people do. With time, a pattern begins to emerge through their searches and re-Tweets, and this was what was manifesting as a personality. In this sense we've successfully created skinny but still potent 'replicants' of our own, running through the streets like in Blade Runner - if not physically then at least digitally - the rain pouring around them, neon lights glowing, living other people's memories.

What we're not trying to do is create mechanical robots. Hardware is a different question. We're interested instead in the totem as an emergent personality; a spirit guide as a digital buddy rather than a clunky, lumbering thing.

When it comes to pulling data for these Weavrs, as we call our bots, to play with, we use many sources but never Facebook. Every time someone mentions internet privacy and Facebook in the same sentence it seems that everything grinds to a halt and people get scared all over again. We clearly have to be sensitive as to how the rest of the world perceives privacy. Do we want DNA data, for example, passing through the hands of Weavrs? Damn right. But it entirely depends on what personal information people are willing to give away.

In The Matrix, characters are able to insert a jack into the back of their heads and directly download information about anything - a helicopter flight manual, for example. With Wikisplashes, a function we developed with the Brainjuicer team, we can do just that for Weavrs. All the hundreds of people who have put together the Wikipedia page for something like fishing, for example, can be instantly synthesized down to a single nugget. And suddenly the bot looks up and says, "I know fishing."

But then things got weirder. It turns out that if you leave a bot like that alone for a few days, an actual personality begins to emerge.

Little tools like this allow us to offer our work as licensed products, giving brand managers and researchers the ability to create quick, virtual embodiments of their target demographics and watch the results weave a reality they can draw insights from - and all without pestering another human being with a single bloody questionnaire.

Away from this, Weavrs also represent the product that takes us back to our storytelling roots. Users can easily program their bot with a range of core characteristics. They can give it interests (photography, beer, mountain-climbing) and emotional associations (when happy they like the cinema, when sad they prefer parks); a routine so the bot knows what it should be doing in the morning, afternoon and on the weekends; and a couple of core locations via a Google Map widget. Then they just press play. Instantly the Weavr will start blogging, producing a half- dozen posts within the first seconds of its existence and continuing throughout the day, going to bed at about midnight and then getting up again around 7am.

As a storytelling framework for these Weavrs, we've started taking on Joseph Campbell's 'monomyth'. It's a theory of what all stories consist of, also called 'the hero's journey' - and if you've ever seen Star Wars or The Matrix or that one with the Hobbits, they all essentially tell the same tale: the monomyth. We've now turned its 17 section, three act structure into an equation that our Weavrs can live day by day, telling an epic tale based upon who they are.

This then gets us into a rather interesting area - and one we're very aware of since it can understandably freak some people out when they discover that there are a dozen Weavrs navigating Silicon Roundabout in Shoreditch, carrying out some kind of epic, semi-fictional storytelling. Clearly they can't be seen. But when you hold an Augmented Reality application like Layar up to the street and you see five Hobbits hanging around the street, it's kind of weird.

But that's also what we like about the whole thing: the questions that this work is raising are brilliant for rethinking how we use the web and what the web is for. It opens up a whole territory of weirdness to explore. We've helped plant the flag of discovery in this land and now we're inviting everyone in.

The next step for Weavrs is an even more curious one. At the moment the bots are making stories out of the data they find. But what if we could give it physical form? What if we could get them to hammer away at a virtual cube with all the pure, raw data that they're already collating, and then 3D print it as jewellery or a sculpture? We've got a long way to go until we get to that level of automation, but the tools for production, like Blender and the MakerBot, are already out there. What we want eventually is a Weavr who not only prints objects but also sells them. The process of pushing these design fictions to platforms like eBay is a very simple one... the question is what to make?

Obviously there's a problem with Weavrs understanding what they produce. But when they share their creations through their blogs, then the audience can decide what is good and what is not and thus affect future production. Traditionally, iteration in product design can be done with sheer force. Google deals with search and retrieval like this, for example, and it does so fairly effectively. There are people out there already using data to generate two- dimensional art. And there's a boutique in Brussels that's just opened which specializes in fabbed art, everything being laser-cut or 3D printed. But getting Weavrs to shape the actual design fictions themselves, that's something new.

If we just have an injection of code, software and data into audience-approved 3D models, then we're getting somewhere close to where product development could be heading. It just doesn't involve that many humans. But maybe it shouldn't involve that many.

Are we really talking about a physical world here populated by digital spirit guides? Sometimes the only language that fits what we're doing is one of mysticism and magic. But when you watch a Weavr blog itself into existence, it's as close to digital magic as anything else I've seen for a very long time.