We're at the beginning of something better
Most talks about AI and media are eulogies. This one, which I gave to journalists at the fjum AI Media Day in Vienna and have edited and translated here, argues the opposite: Right now is the best moment in living memory to go independent and start a newsroom.
Almost every story we tell about media runs in one direction: It keeps getting worse. How long can we hold on?
If you’re young, you don’t know if you’ll make it into a newsroom job. If you’re older, you’re wondering if you’ll make it to retirement. Either way, you’ve been hearing it for years.
I believe the time has come to leave this story behind, and the self-fulfilling prophecy it has turned into. Today, for the first time in a long time, we have a real chance to build something more creative, more interesting, and more resilient than the old system.
Let me make the case in three steps.
1AI is getting better, and it won't stop
We can keep this point short and simple, as it’s blindingly obvious to anyone following AI: The models are improving rapidly, their capabilities keep climbing, and there’s no good reason to think it will stop.
Anthropic
OpenAI
GoogleThis is real, although there is a trap some folks fall into that makes them think otherwise. It usually goes like this: They tried ChatGPT in 2024, decided it was a cute toy, and put it down.
That was a different world. We’re not in it anymore. This stuff works now.
2The system we have is fragile
I want to be precise about that word. From outside, the industry looks embattled, which is not the same as fragile. Embattled is having a hard time. Fragile is having no solid ground under you. My claim is the second one: Almost nothing holding this system up is something we control.
Start with the money, because that part isn’t controversial. The commercial floor is falling out: fewer readers, print bleeding, the money sliding to digital where it’s thinner, and the ad revenue that paid for journalism going to tech companies instead.
The audience is aging, the under-25s who’d replace them mostly aren’t there, and the next blow is already landing. People ask a chatbot and never come to our site, and we cannot monetize a reader who never shows up.
It gets worse, because a lot of what we do has stopped being worth anything. Be honest: a huge chunk of our daily output is taking someone else’s reporting and reformatting it, re-headlining it, repackaging facts everyone already has. That was never much of a business, and now AI does every bit of it in seconds, for free, in any style you ask.
Jay Rosen has a name for the journalism this produces: the view from nowhere. It is reporting with no point of view, the kind that sits between all sides so no one can call it biased. That is the easiest thing of all for a machine to copy. The value of that work is drifting down, and will go to zero.
So the money is going, and here’s the fragile part. Every solution we reach for has the same shape: the creation of dependency. The first is public money. When the market stops paying, we lobby for subsidy. In some places it’s working, especially in Europe. But be clear about what it is: total dependence on a single bet, that politics keeps caring about funding us on today’s terms, indefinitely. Governments change. A new one cares less, or notices how much leverage that money buys it, and someone other than us decides where it flows.
The second is en vogue at journalism conferences: Make the tech companies pay. It used to be “Facebook should pay for our videos”; the 2026 version is “make OpenAI pay to train on our articles.” I fear that’s a dead end. They don’t need us. A startup that licenses training data told me the labs don’t seem to need more, certainly not text. They train differently now; individual articles aren’t valuable. And even if the money showed up, it’s the next dependency. Same trap, different landlord.
To be clear: None of this is for lack of trying. There is excellent journalism happening in existing newsrooms all over the world. Some of these newsrooms really want to transform, and a few will pull it off. But at that scale it comes down to leadership: someone willing to reinvent their organization technologically as well as culturally, and to bring everyone along. Too often, that leadership is missing.
So that’s the diagnosis. The thing that pays is collapsing, and every substitute is something we don’t control: a market drying up, a state that can change its mind, a tech industry that doesn’t need us. That’s what fragile means. If those were the only options, the pessimists would be right.
They aren’t the only options.
3The new thing is better, and it's coming
Think about it the way a market would: What’s becoming scarce?
The cleanest way I’ve found to reason about it is almost technical, and I owe the framing to the writer Jasmine Sun: What is there no training data for? Because anything there’s training data for, an AI model will eventually learn to do. What’s left is what only a human, in the world, can produce. Three things stand out:
- What isn’t public yet. Your own sources, being somewhere first, being trusted, being in the war zone, having the document nobody else has. Investigative work. By definition, there’s no training data for a thing that hasn’t surfaced.
- The real world. Being in a room with people, live: a talk, a podcast, an event, a community. What I’m doing right now, standing here, is not something a chatbot can do. And in a world flooding with synthetic everything, I think presence is worth more, not less.
- A real opinion. A clear stance, a point of view, charisma. Surprise, weirdness, unpredictability. There’s no training data for what a specific human actually thinks and feels.
I know it’s hard to tell a company that its future is charisma, but it’s actually a useful frame. One funny recent example comes from none other than the pope. On his trip to Spain, he was asked which football team he supports:
Reporter
We're going to Spain. The big question is: Real Madrid or Barça?
Pope Leo XIV
Oh, that's easy. The Pope is for all teams. But Robert Prevost is for Real Madrid.
Robert Prevost is the pope’s birth name. Ask yourself: What part of that answer could ChatGPT have written? And what part is human?
The authentically human lane is wide open, because incumbents tend to underinvest in it. It already works, too, at every scale. In the US, Dwarkesh Patel turned a one-person podcast of deeply researched interviews into one of the most respected shows about AI and science. In a tiny market like Austria, Andreas Grassl built an audience explaining world politics on TikTok and Instagram, and now makes a living from videos, podcasts, and event appearances. If it works for them, it works for others.
You don’t need a mass audience. You need a sharp niche where you’re the best by a distance. Find a thousand or so people who care about the precise thing you care about, willing to pay a few dollars a month, and the math starts to work. And that’s the traditional layer. As a next step, you might consider B2B: working with companies that turn out to be very interested in understanding the world.
The new newsroom is two people
The obvious counter-argument: Well sure, but aren’t these creators just repackaging existing information again? Who will do real investigations?
I think this will work! Picture a newsroom of two. One person does deep research, and I don’t mean a story a day. I’m thinking more like one story a month. The good stuff, investigative or civic reporting that has been lost in so many places. Radically human, out in the world, talking to people, digging through data, finding things.
That reporting gets fed into the database, with AI agents doing the connective tissue: transcripts, data analysis, fact-checking, finding angles, formatting. The business runs the same way. AI tools collapse the cost of the back office that quietly sinks small outlets: bookkeeping, contracts, marketing, coding, sales, admin. What used to kill scrappy newsrooms was needing eight salaries when you could afford two. Two can do it now.
Importantly, the humans also do the part that matters the most: the telling. The writing, the voice, the presence. That’s the part with no training data. That’s us.
You publish the story, and then you don’t shut up about it. You go on every podcast and YouTube channel that will have you, speak on every stage. You find your people on these platforms, and then work on moving them to channels you own, like newsletters.
So who pays for it?
Start with the audience we already know: humans who trust us. The familiar tools work, subscriptions and memberships. But so do plenty of things we don’t usually count as journalism: consulting, building and selling tools, live events. Here are two entrepreneurial examples:
- The media company Every bundles daily essays with the AI tools it builds and sells, all in one subscription.
- The journalist Sophia Smith Galer built Sophiana, an app that helps turn reporting into social videos, first for herself and then as a product anyone can buy.
Then there’s a second audience, only starting to form: agents. Not the scrapers that train the big models, but AI that reads and acts on a person’s behalf. Soon they will spend money. Imagine logging into ChatGPT and getting access to what a specialized outlet knows. That’s valuable to businesses looking for an edge, and journalism can provide it.
The agent economy is also where micropayments finally start to make sense, not because humans will ever tolerate them but because you can hand an agent a budget and let it pay a few cents per read, thousands of times over.
What media funding should actually do
None of this pays on day one, though. The first year, before you have the audience or the products, is the hard part, and it’s exactly where funding and philanthropy should come in. Importantly, this money must be spent on independence instead of dependence.
So, if you’re in the funding space: Fund as many media startups as possible through that first year. We don’t need one; we need hundreds. And fund them seriously enough that founders aren’t building at 2 a.m. around a day job. The good news: This new form of media startup has almost no overhead cost. After a year, the ones that work carry themselves, and the support ends. They owe nothing to anyone.
A politician can call one editor-in-chief of a large newspaper and lean on them. But you cannot call hundreds of podcasters, creators and independent writers. And yes, some of those startups won’t make it. That’s fine; that’s how it works. But the ones that do will be unbuyable. Some funders, like the Media Forward Fund, are already starting to work this way, and I’m proud to be on their jury. We need more of it.
Of course, there are limits. Not everything will pay for itself. Some journalism is just too civically important to lose, and is never going to turn a profit. That work will always need support. But ideally even that funding should be diversified and sustainable, so no single source can pull the plug.
Where the optimism comes from
I know this would all be thin if I were only describing a future. But here’s the thing: This is how Verso works today.
We started with workshops and talks, expanded into building tools, and now run in-depth programs like the fjum AI Media Academy. We haven’t grown the team—it’s still the two of us, and it’s working: The company grows; the team stays flat. The only thing that really expands is the quality of what two people can do, because AI tools keep getting better. We launched in 2024 and hit our financial goal within about a year.
Is giving a talk journalism? Is building software? I’ve never quite understood why it counts on a website or in a video, but not on a stage. The job is to get good information to people, by whatever route the world allows. If the power went out and we all gathered in the city center and someone shouted the news through a megaphone, that would be a journalistic act too. A newsletter, an app, a talk, a piece of software, whatever works. Once you get good at one, you can expand.
None of this means giving up on the newsrooms we have, and helping them make the same leap is a big part of our work. But that conversation is already happening everywhere; this one has barely started.
It’s the best time in the world to start something. The first year is fundable. The tools are good enough and the landscape is competitive: many companies serve them, the open models keep getting better, and you can even run them yourself. You’re captive to no one. The lane is open. If you’ve ever thought about building something, this is the moment to try.
And if you want to talk it through, that’s literally our job. Reach us at [email protected].