Meredith Kopit Levien runs The New York Times. And while every media company is figuring out how to survive AI commoditisation, she's betting on something unfashionable: human expertise you can't fake.
In a conversation with Ben Thompson at Stratechery, Levien laid out the Times' strategy - and it's not what most publishers are doing.
No pivot to AI-generated listicles. No content farms. No chasing algorithmic distribution. Instead: recipes tested in actual kitchens. Puzzles crafted by human constructors. Journalism with rigorous verification.
The thesis is simple: when AI can generate infinite mediocre content, the moat is quality you can only get from people who actually do the work.
What Makes a Recipe Worth Paying For
Here's the bit that stuck: NYT Cooking doesn't just publish recipes. They test every single one. Multiple times. In real kitchens. With actual feedback loops.
An AI can scrape a million recipes and generate variations instantly. But it can't tell you that the oven temperature needs adjusting for fan-assisted ovens, or that the dough feels wrong at step three, or that the timing assumes your vegetables are at room temperature.
That tacit knowledge - the stuff that comes from repetition and failure - is exactly what AI training data misses. And it's what subscribers pay for.
Same principle applies to Games. Wordle wasn't algorithmically generated. Neither are the crosswords. Human constructors bring wordplay, cultural references, difficulty curves that feel right. You can't A/B test your way to that.
The Journalism Bet
Levien also addressed the elephant in the room: the Times is suing OpenAI and Microsoft for copyright infringement. Their argument is straightforward - training AI models on proprietary journalism without permission or compensation undermines the entire economic model that funds reporting.
But the more interesting point is strategic, not legal. If AI can summarise articles, why would anyone click through to the source? The Times' answer: because summarisation isn't journalism. Investigation is. Verification is. Access to sources who won't talk to chatbots is.
Commoditised content gets summarised away. Original reporting remains valuable because you can't generate it from other people's work.
Sports, Video, and Community
The interview ranged across the Times' expansion into sports coverage (The Athletic acquisition) and video production. The thread connecting them: expertise and community, not distribution.
Sports coverage works when writers actually understand the game, know the players, have relationships with teams. That's not something you can prompt-engineer. Video works when it's crafted, not just uploaded. Community works when people trust the source enough to engage beyond consumption.
Levien's framing positions the Times not as a newspaper fighting AI, but as a platform for human expertise in an era when AI makes expertise more valuable, not less.
What This Means Beyond Media
The Times' strategy has implications for anyone building in an AI-saturated market. The pattern is: find the work AI can't replicate, then make that your moat.
For recipes, it's kitchen-tested reliability. For puzzles, it's human-crafted delight. For journalism, it's investigative access. For your business, it's probably something similar - tacit knowledge, relationship-based work, judgement calls that require context AI doesn't have.
The risk of this strategy is obvious: it doesn't scale like AI does. Testing recipes is expensive. Training journalists takes years. Human expertise is a bottleneck.
But Levien's bet is that bottlenecks become moats when everything else is abundant. If AI makes content infinite, scarcity shifts to quality. And quality still requires humans who know what they're doing.
Not every business can follow this playbook. But for those who can, the message is clear: the answer to AI commoditisation isn't fighting the technology. It's building something the technology can't copy.