Change Agent

The journey of a thousand miles begins…with having some idea of where you ought to go. And a desire to go there.

Perhaps that’s not as inspiring or pithy as the original Chinese aphorism, but it’s probably more apt for the AI age. (And pithy is overrated, in any case.)

I confess that I only learned the phrase, “theory of change,” late in my career, after I became more immersed in the world of non-profit journalism and philanthropy. It’s a concept that makes sense, even if it’s been parodied to death and is hopelessly overused; but the notion that you should have an idea of how some intervention you’re making will lead, eventually, to some desired outcome, is not a bad one.

For years it wasn’t something you thought about in for-profit media; not least because your theory of change was: Are we making more money, or not? It wasn’t complicated. And more to the point, at least during my formative years in print newspapering — ah, those sweet, uncomplicated days of my youth — there was only one real business model, one real theory of change, and we all more or less followed it.

And now: Well, good luck articulating what the world looks like next year, let alone a strategy for success in it.

But it may be the most important thing we can do.

I was reflecting on this at the Nordic AI in Media Summit in Copenhagen, where a host of smart speakers (Ezra Eeman! Shuwei Fang! Nikita Roy! Florent Daudens! and more!) sketched out the multiple upheavals they saw coming in the media landscape. And yet, I was thinking, almost every AI tool I’ve seen demoed or deployed so far — even the best, most creative ones — is built around optimizing for today’s environment and today’s newsroom. Our best work — and I include (as I would) Semafor in this — is designed to help solve today’s problems, not position us for tomorrow.

That makes sense: We need better efficiencies and capabilities right now, and expending energy and resources exploring and building tomorrow’s AI systems might not be much help in the moment.

But if not now, then when? At the pace AI capabilities are advancing, we’re likely to wake up one day and find the standards and systems governing news and public information have already been embedded into the technology infrastructure of the landscape, by people who don’t have the public’s interest at heart. We shouldn’t cede that space; we should at least be in that discussion.

And it starts with some idea of what we think the future looks like — a “theory of change” — and ideas about what strategies, interventions or systems will make that a better place for the public, and ideally for us as well.

Absent a vision of the future, it’s hard to imagine what we could meaningfully build; and that leads us by default to building for what we know, which is the world of today. And that’s great — for today.

Which is why we need to be engaging much more actively in debates about what that new world looks like, and building experiments that test solutions and ideas for our visions. To be sure, there are already some underway: Shuwei and David Caswell are developing frameworks for investing in new news infrastructure, and Sannuta Raghu’s News Atoms is a fascinating experiment in creating a new information architecture. Florent is trying to create an agentic newsroom. But they’re a minority.

There’s an understandable hesitation to put bets on ideas that may yet turn out to be wrong. No one wants to burn resources backing the wrong horse. But not making a bet on the future is another kind of bet — that the revolution isn’t imminent, or that the new world will look a lot like the present world.

You can hear that thinking if you roam the halls of any journalism conference. Get the platforms to pay for content. Do more original journalism. Build deeper relationships with audiences. Drive more direct traffic. But those aren’t theories of change; they are theories that the world won’t change that much, and that the strategies of the past will serve us well in the future, if only we execute them better, faster, and cheaper.

For what it’s worth, here’s my vision: That the vast majority of people will turn to agentic systems that offer them relevant, personalized information, when they need it; and that a key question is whether that information is created for them with their interests in mind, or whether those agents are working for someone else. In that world, we’ll need to find ways to structure and standardize our information and data, and ensure that some level of verification and provenance travel with that information; we’ll have to get to know our users and their needs much better, in depth, and quickly. We’ll have to see if we can create agents that can scour the web and our own data to find accurate answers to them. And more.

I could be way off base. But it’s a theory of how the world will change, not how it won’t.