I’m greeted in the lobby by a man dressed all in retro steampunk garb, tweed and cogs. He is sporting a moustache that demands the addition of a pipe. Going by his family name, Haldane, and with the title ‘Delivery Coordinator’, he leads me through to some open office space, through a door labelled ‘The Monkey Pit’ written in what look like an array of little pistons raised or lowered, acting as pixels. As we pass through, I’m sure I see one move with a little puff of steam.
The people working at the Code Zoo don’t call themselves coders. As I’m told by Maggie, one of the apprentices, “We don’t develop apps or dynamic content. We growing and harvesting living code.” This code is called G-bots. “What we do isn’t coding, it’s more organic than that.” I nod, but secretly I am sceptical. If you work with code, you’re a coder, right?
Maggie is just one of over 15 people who together form the Code Zoo collective full-time and about half again on an as-needed basis. The company itself is one of the new wave of self-owned entities, a model that is increasingly being promoted by IMF2 as part of its strategy to foster re-globalisation without the stranglehold of multinationals that triggered the world-wide IP crisis.
“The self-owned entity approach fits our dynamic very well” says Haldane, removing and worrying his pince-nez, genuine antique, not printed. “We trade internationally and need the protection afforded by company status, but do not want any of us to actually ‘own’ the business as an entity. It is such an arcane concept. Also, it keeps everybody focussed on the bottom line – no profit, no pay,” he winks. Each employee receives a share of the company profit each month, in place of a wage.
About a third of the collective work at this site in Toronto. The others are spread across two other offices, one in South Africa and the other in New Zealand, and then a few others working entirely remotely. “We have clients in all time zones, so we need people who are in the office at all hours.”
Contrary to what you may think, Code Zoo takes off-time very seriously. Apparently working hours are strictly limited, as they have found that performance drops off radically otherwise. “It’s cheaper to split the working profit more ways than carry someone who is underperforming because they are tired or need to spend some time off with their family, or just need to relax for a bit.”
The first person Haldane introduces me to in the monkey pit is Sandy, a serious haircut in her mid-twenties. Sandy describes herself as a “herder”. The company harvests G-bots from the wild and grows it up on dedicated hardware, breeding and selecting it for sale.
“A code herder”, Sandy tells me, “is responsible for moving running G-bots between different hosting providers. Our major cost is CPU time,” Sandy explains. “We have to constantly move code between data centres to take advantage of lower unit pricing.” Why can’t this be done by software? “It’s not just a case of chasing daylight hours in different time zones. It depends upon local holidays, price spikes caused by big sports events, even weather.” The weather? “Yeah. The major cost of hosting is cooling the data centres. Cold weather means less air conditioning means lower unit cost.”
Also, she tells me, you can’t just move any code anywhere. They form ever-changing community structures which work more efficiently when co-located. Splitting a community across the globe could render them non-functional, and therefore, their services unsalable. Ironically, there is currently no algorithm that can optimise this problem, so a human makes the judgement calls. That human today is Sandy.
Next to Sandy is a large display. It takes up several cubic meters of the room, and within it is a tangle of lines and blobs that looks a lot like a social network graphs. “That’s pretty much what it is, but those aren't people, we’re looking at the G-bot community structures. Lines between them show the messages they are sending and recieving.”
On the other side of the office there are three people who so far have not spoken a word, immersed in goggles. Haldane taps on his phone, and one of these de-goggles and turns around. After a moment distractedly easing numerous rings back onto his fingers, he introduces himself as Mark, one of the G-bot tamers. His role is to domesticate wild code, encouraging it to perform useful functions. This is done through a mixture of direct intervention and evolutionary pressures. “The trick is to tinker enough that they want to do jobs for us, but not design their innate adaptability out of them.” But why bother with wild code in the first place? “The code that lands on my desk is raw. Every G-bot is unique, doing different things in amazing ways. We could never sit down and design all of this by hand, and we simply do not have the cycles to evolve this diversity from scratch.” Hence tamers are needed, to weed out traits that prevent them playing well with others, and cataloguing what they may be useful for later.
“It is a really rewarding job. You never know what you’ll find, and there’s an art to getting G-bots ready for commercialisation.” The tamer team currently works to a 1:20 success ratio. “It doesn’t sound like much, but each success goes into the zoo and may be sold multiple times. The 19 of 20 that don’t make it that far are still very useful for us, shaping our internal evolutionary environments.” When I ask Mark more about commercialisation, he tells me that it’s not really his thing. “I just tame them.”
It’s now after lunch in the Toronto office. I’ve just spent an hour attempting to re-design my own G-bots starting with some from the zoo. Real hello-world stuff, a simple spam filter. It is enough for me to see the potential and to know that it’s a job best left to the professionals. My bots keep finding loop-holes in how I’ve described what I’d like them to do, taking the laziest way out rather than actually doing what I want. “Sneaky, aren’t they” Sandy calls from the other side of the monkey pit. I have to agree.
Over mid-afternoon coffee down the street, Haldane and another of the collective, Ada, a “feeder”, describe how G-bots or G-bot communities are productised. “It’s all a question of “feeding” them with the right kind of questions. We evolve them in environments where the population have to solve these to survive and reproduce.” This is similar to how I was growing a spam filter, but instead of feeding them my inbox, the Code Zoo build G-bot swarms that model everything from election outcomes to how to route physical goods across transport networks.
Ada takes a long pull on her tall coffee and goes on. “There’s an art to it though”, she says. “You have to keep giving them new examples or they have an annoying habit of just memorising what they’ve already seen and then act all clueless when given something new.” Haldane interjects, “For each client, we actually have a dedicated data-collection team who build the ‘food’ data. In fact, part of my role here as Delivery Coordinator is to understand what kinds of data are needed to solve the client’s problem.” “And then we make sure that this is fed to the G-bots, so that we have the best chances of them becoming product” continues Ada, as she drained her coffee.
Back at the office, and my day at the Code Zoo is nearly over. I’m sure that the “monkey pit” room sign looks different than it did before – perhaps a different font? I’ve enjoyed my day here. I’m still not quite sure how they manage to encourage the G-bots to behave – I couldn’t really make mine do what I wanted, it really is an art and bares about as much resemblance to coding as printing clothes does to knitting, but their company results speak for themselves. Original pince-nez are going for more than the price of a small family home, with provenance.
Maggie is leaving the office at the same time as me, and as we walk to the mass transit I ask her about the future. “This sector is definitely a great one to get in on. Perhaps in a few years much of what we do will have been automated, but that will just create new roles, won’t it.” She seems committed to the industry even if at some point she leaves the Code Zoo collective. “G-bot based services are here to stay, and so am I.” I expect she’s right, on both counts.
