Dylan: Is Nick Lichtenberg the future of journalism? In other words, are we still in the “power tool” phase, where A.I. is essentially a productivity enhancer, or are we already staring down the barrel of a full-stack disruption of reporting and writing, itself?

Ian: It sort of depends on the business model. For the top writers, we’re nowhere near the point where the content is better. The Lichtenberg approach is much more radical than how tech journalists like Alex Heath and Casey Newton are reportedly using A.I. to help with transcriptions, research, and even drafting copy—you know, the standard-ish optimizations. For media companies that rely on traffic—clicks and page views, etcetera—there will be a real temptation, maybe even a business necessity, to optimize for higher revenue and lower costs. After all, A.I. is almost the perfect tool to feed that impulse for more, faster, cheaper, and hookier content. I don’t think it’s crazy to imagine a world where A.I. incentivizes some startups or diminished legacy brands to take another shot at the content-farm model, even if it depletes user trust.

Dylan: So is the real story here about individual journalists experimenting with A.I. to give themselves more leverage, or institutions quietly redesigning their workflows to lower costs?

Ian: It is absolutely the latter. Heath’s approach to his work—as detailed in that Wired piece—may surprise people. He noted that he begins writing, for instance, by dictating his story into Claude. But Heath is making a judgment call, and he has a ton of credibility and a remarkable track record to back it up. The process works for him, and his audience doesn’t seem to mind if his work is A.I.-supported. Most importantly, it’s great that it’s his choice to make. Good journalism isn’t cheap or easy, which is one of the reasons the industry is in trouble. Maybe there’s a reason to be optimistic that A.I. can help lower some of the back-office costs—admin, legal, marketing, web services, etcetera—and allow small businesses and operators to focus more on the journalism itself.

But you can imagine how scaling that approach, with lesser talent and less discernment, would be bad for the industry writ large and our information ecosystem. More importantly, I remain worried that large news organizations will be tempted to return to the worst instincts of the S.E.O. and social media era.

Dylan: Which parts of the journalistic process are most vulnerable to automation: reporting, writing, editing, distribution, or audience targeting?

Ian: There’s a difference between what we can do right now and what we should do with these technologies. From a sheer capability perspective, pretty much everything that doesn’t include picking up the phone or meeting a source for coffee is vulnerable to automation. It’s kind of like self-driving cars—the cars today are definitely good enough to drive themselves, but everything gets more complicated when you start dealing with edge cases and judgment calls and risk allocation in the real world.

Anyway, some of these things are easier than others. Audience targeting is easy for A.I., and automation has been accelerating in that domain for years. Journalism is more complicated. Large language models are quite literally designed to generate text. Does that mean they’re good writers? I would say no. Since they were trained on such a vast trove of writing, can you ask an L.L.M. to edit a piece along certain guardrails? Also yes. Will the result be better than not editing at all? Depends on how clean the writer is. Will it be better than a human editor? Almost certainly not. Are they good for fact-checking? Definitely not. Could they point a reporter in different directions? Maybe, so long as the reporter genuinely understands the vulnerabilities of the system they’re dealing with.

A nontrivial concern of mine here is not just in potentially offloading cognitive work, but in subtly beginning to cede decision-making and judgment calls to products owned and operated by the large technology companies. Silicon Valley has already put a lot of money and effort into owning and controlling the collective conversation—think of all the social and legacy media platforms owned today by tech billionaires. I’d be wary of giving them another lever of control.

Dylan: If A.I. dramatically lowers the cost of content production, what happens to the economics of media? Does it hollow out the middle and concentrate value at the very top?

Ian: If readers actually want more rather than better journalism, then you can imagine a world in which the cost of content production essentially becomes equivalent to the cost per token charged by OpenAI or Anthropic. But the reality, of course, is that that’s not what most people want. So there’s going to continue to be a floor cost to produce quality journalism. The real question is, how much cheaper can newsrooms get without sacrificing their brand, mission, or, most importantly, audience? That’s a much harder question to answer.

Dylan: What does this do to the newsroom itself—especially entry-level jobs and the traditional pipeline for developing reporters?

Ian: They’ll likely keep getting smaller. Entry opportunities will get even thinner on the ground—but again, this is not a new problem. The traditional pipeline died when local news died.

The silver lining is that there’s a world in which the cost-savings would allow more companies to invest more in meeting market demands, like local news, rather than racing A.I. to the bottom. In this way, A.I. and the enshittification of everything is potentially a massive opportunity for new orgs to pop up and new (old) pipelines to be established. I am not without hope here, but achieving that future takes money, time, and ultimately conviction—not just in media as a business, but in journalism as a service. That’s gotten rarer as time goes on.

Dylan: Are licensing deals going to save news organizations? And who on the media side is navigating those deals the best?

Ian: I think it’s dangerous to play for short-term Band-Aids when the problems ahead require long-term approaches. That said, I don’t know that anyone is doing it better than anyone else; data is absolutely vital for A.I., especially high-quality data. I think a lot of brands got in bed with the startups too soon, before they understood the value of their data, and before the fundamental I.P. questions were answered by the courts. In 2023, OpenAI had the fastest-growing app in history and all the cards; in another year or two, that might not be the case. But, in the words of Tom Petty, the waiting is the hardest part.

Dylan: Five years from now, what will feel most different about journalism—and what, if anything, will prove surprisingly unchanged?

Ian: If my crystal ball were that effective, I would become a V.C. But if I must guess: I’ll start with the audience, which has largely been ignored in this debate. The demand for high-quality, trusted journalism will continue to grow, especially as more organizations and platforms take the easy, clicks-first, A.I.-aided approach and A.I. continues to proliferate—further reducing our capacity to trust what we see online. The people and organizations delivering on that demand will, even five years from now, not have automated that process of journalism. We’ll see A.I. being used when it makes sense—the way we email or text sources today whom, 30 years ago, we would’ve called on the phone, or the way we spend 10 seconds on Google instead of a few hours in the library. We’ll have another tool in the arsenal, a tool that’s great for data analysis, quick transcription, and targeted web searching.

I think it’s fair to say that real reporting—source building, researching, conversations, relationship building, trust—will continue to get rarer. Fewer organizations will invest the money or time in that process, but the ones who do will win.

See you Friday,
Dylan