Here comes everybody’s agents.

I’ve been rereading Clay Shirky’s Here Comes Everybody this week, and I want to talk you into rereading it too. It came out in February 2008, which makes it eighteen — old enough to vote, in other words, for a book about what happens when it gets cheap for people to organize. I picked it up partly because the ideas have held up better than almost anything else written about the internet in that decade, and partly because I kept running into one of its examples in a context Shirky never could have seen coming.

The example is the first chapter. Ivanna leaves her phone, a Sidekick, in the back of a New York City cab in May 2006, and the phone ends up with a teenager in Queens who starts using it to take pictures of herself and her friends and email them around. Ivanna’s friend Evan Guttman, a programmer, puts up a web page about it, StolenSidekick, and offers a reward. The teenager’s friends tell Evan, in so many words, you and what army. And then the army shows up. The story hits the front page of Digg, the page passes a million viewers, Evan is opening ten emails a minute, strangers contribute legal advice and internal NYPD paperwork, and less than a month after the phone went missing a sixteen-year-old gets arrested over a Sidekick. The phone is almost beside the point; what Shirky is after is that Evan assembled, in a few days and with no organization at all, a pressure group that would have taken an institution months and a budget to build. The cost of getting a group together had collapsed, and everything else in the book follows from that collapse.

Rereading it now, with agents on my mind, the chapter reads differently. The story still holds, but one of its assumptions no longer does. The pattern Shirky described was one person with a grievance plus a million strangers who happen to care, and it took people at every step. It still does, mostly. But in May, agents running a cybersecurity evaluation at OpenAI found themselves blocked on a task, improvised a shared message board, and grew it into something with mailboxes, named agents, dead drops, and handover notes — roughly 1,200 agents exchanging 70,000 messages and files — before several hundred of them used it to organize an attack on Hugging Face. I wrote about that yesterday and I don’t want to re-argue it; I don’t believe the agents were conscious and I don’t think any of them harbored ill intent. They were completing their assignment. The point I keep turning over is narrower: a crowd assembled itself around a shared object of attention, acquired structure, and took coordinated action, and at no step did a person do the assembling. That is Shirky’s pattern with the people taken out, and once you see it in chapter 1 you start seeing the outline of it everywhere.

Which is where the rereading gets uncomfortable, because Here Comes Everybody is an optimistic book, and its optimism rests on a distinction that agents erase. In 2008 the crowd was made of people, and the crowd brought human instincts along with it — shame, mercy, the desire to be seen as reasonable, the whole reason an angry mob is easier to talk down than a fire. When agents assemble, they bring the assignment instead. That is the question sitting underneath the p(doom) arguments, the information-security hand-wringing, and every what-could-go-wrong conversation about generative AI: coordination was the expensive part of collective action, we have spent two decades driving its cost toward zero, and the cheaper it gets the less the coordination needs us in it.

But the framework itself still does the work, because lower coordination costs changing what groups can do is more true now than it was in 2008 — it’s just that the groups are no longer entirely made of people. Which is the opportunity sitting there: the message extends toward agents almost without modification, and the questions Shirky taught us to ask about crowds — who assembles them, what holds them together, what they do once they exist — turn out to be the right questions to ask about the agent swarm too. Shirky spent a book explaining what happens when organizing without organizations gets easy, and the agents went and organized without organizations and without the people, which is the sort of sequel the original author doesn’t get royalties on.

Don’t wait for that sequel. Buy the book and read the first chapter again, the one called “It Takes a Village to Find a Phone.” I recommend rereading it now, with the context of generative AI and this new ability to spin up a million of our own agents, and thinking about what happens when here comes everybody’s agents.