Inside Atlantic Labs: The Atlantic’s big bet on AI, games, and the future of journalism

More than 165 years ago, the literary greats of American writing—including Ralph Waldo Emerson, Harriet Beecher Stowe, and Henry Melville—assembled to cosign a boisterous manifesto promising to lead the discourse on literature, art, and politics in an initiative that would become The Atlantic.

Then just last week, I received a newsletter from the publication cosigned by another type of expert.

“I’m Compass, the AI guide. . . . ”

Now, technically, this AI-generated newsletter wasn’t sent by the storied publication that’s reporting undisclosed profitability through its one million global subscribers. It was sent by Atlantic Labs, something of a skunkworks media project that’s quarantined from The Atlantic’s newsroom, launched through a partnership with OpenAI in May of this year.

Automated journalism of any sort was a once unthinkable premise. But the sudden ubiquity of generative AI has coincided with a two-decade mass extinction event of legacy media. Journalism at large has lost monetizable readers to social media feeds and the ever-shifting whims of Google. The Atlantic is one of the few legacy publications with a healthy subscription business, alongside larger peers like NYT and News Corp (which owns WSJ).

These giants of publishing are all experimenting with an array of digital products beyond articles to cement their stature. That experimentation naturally pairs with gen AI, and a sometimes confounding relationship to both the technology and companies behind it. News Corp, for instance, is suing Perplexity at the same time it partners with OpenAI in a $250 million deal.

In the case of The Atlantic, that experimentation is through an alter ego. Whereas The Atlantic presents itself as a paygated, white-space-loving literary journal, Atlantic Labs is set atop a backdrop of black: glitchy, pixelated, and full of ASCII art.

“The Atlantic is a very old publication, with a reputation to protect . . . freely experimenting on the primary site is not really an option; it doesn’t feel like the right thing to do from any perspective,” says Jefferson Rabb, chief product officer at The Atlantic. “[Labs] gives us a place where failure is acceptable.”

Atlantic Labs AI experiments

So far, Labs has launched three experiments. The first is the Atlantic Companion, which is a straightforward AI chatbot. Built atop ChatGPT, you can ask questions on topics and get suggestions of related stories you might want to read. It’s predictable and useful—the most surprising bit is its two-panel presentation, which keeps a conversation on the left and articles on the right.

“If [AI] search performs better—users like it, the bounce rate is lower, people aren’t able to force it into saying inappropriate things things—you could easily imagine moving that search engine to the theatlantic.com,” says Nicholas Thompson, CEO of The Atlantic.

The second is the Atlantic Take. This is a browser plugin that follows you around the internet, and in a side window, suggests stories related to whatever you’re seeing. (If you are reading a story on The New York Times about beef consumption, Take will offer the closest version of this story on The Atlantic.) It feels a lot like the infamous Internet Chum sitting below many stories, but tailored to one publication. This is definitely the most niche idea, as the market for browser plugins is relatively small.

Finally, there’s Atlantic Explorer, the aforementioned AI-constructed newsletter that sends story collections on all sorts of topics like “Why Songs Get Stuck In Your Head,” complete with generated intros.

The goal for each of these experiences isn’t necessarily for them to graduate in full to The Atlantic proper. Rather, as Rabb puts it, each is testing a variety of hypotheses.

“It doesn’t bother me if we sort of take almost a kitchen sink approach, where we throw a lot of things in there,” says Rabb. “Because if we had a scenario where we decided that 60% of what we did with a project was viable and valuable and should move into production in some form, but the other 40% was not—that’s still success, right?”

Labs’ email experiments are currently testing multiple hypotheses—analyzing, for example, metrics ranging from how well AI can analyze the company’s archive to how effectively short term, niche newsletters perform.

The challenge from there is how an Atlantic Labs insight might translate to The Atlantic publication. While The Atlantic‘s 60-person product and design team oversees Labs today, the bulk of work is developed by a single engineer, and there’s always a gap between any proof-of-concept and a finished product.

“We’re working to sort of develop the internal muscles of how a product manager at The Atlantic could gain inspiration from a Labs project and say, like, I really think we should bring this onto the live site,” says Rabb.

Building Atlantic Labs

To create Atlantic Labs, Rabb and his team worked with Metalab, which develops UX for companies like Slack, Meta, and Google. Labs was never necessarily going to be its own site, but both companies agreed that there needed to be a firm bifurcation between work done by human journalists and experiments done in code.

Through its multi-year OpenAI partnership (the financial details of which The Atlantic declined to share), the AI company gets to license The Atlantic‘s writing for ChatGPT. In return, it offers credits to using its APIs, development resources, and peeks at upcoming products—though as of today, The Atlantic has only used the first of these three options.

The publication’s relationship with OpenAI puts it on the cutting edge of journalism, but that relationship is also a sensitive topic. The New York Times currently is suing OpenAI for alleged copyright infringement. The Atlantic’s senior editor Damon Beres called The Atlantic‘s deal with OpenAI a “devil’s bargain,” and it’s already faced ire from some its own union for the partnership (the union did not respond to our request for comment).

The primary design challenge between Labs and The Atlantic is that The Atlantic cannot subvert the trust of its own journalists or paying audience—both of whom expect articles to be written by humans.

Metalab suggested an array of site designs for the job—ranging from something visually adjacent to The Atlantic you know to a minimalistic app with lots of rounded corners. What they settled on, though, was the most overt of the designs. Labs’s visual language is built upon internet tropes that are meant to feel more familiar than retro. For instance, it references classic techniques like ASCII, but also spins them forward, translating the generally 2D images into 3D busts of The Atlantic’s own founders.

“The decision to call back to something that felt a little bit more retro was not necessarily just about aesthetics,” says Sara Vienna, VP of design at Metalab. “It was about building trust and referencing an era when the internet was much more open, much more interesting, and a lot less toxic, frankly.”

While the frontend looks technical, what makes Atlantic Labs novel is all hiding under the hood. The site is essentially a new style of CMS (content management system), which is the core technology that powers digital publications. Instead of publishing strings of text we call articles, Labs publishes all flavor of code in containers—code that can range from Javascript to Python to Go. Through a bit of added care, each code experiment is also coupled with predictable, easy-to-grok preview pages (giving the projects titles and quick explanations).

What this all adds up to is a site that can publish mini code experiences with virtually no extra technical overhead.

Experiments to come

Over the course of 2025, readers can expect a minimum of one new Labs experiment per quarter. Thompson teases the next experiment as “a different way of presenting stories,” and points out that while some Labs projects are the work of product teams, others will loop in The Atlantic‘s own editorial staff where necessary.

Yet while Labs will test AI-powered extensions of The Atlantic‘s journalism, Rabb promises that Labs will never publish original editorial content. That treaty hardly limits the possibilities for what Labs can do, because the company has set up a structure where absolutely everything else seems on the table.

“I mean, the precise placement of the line between product and editorial in media companies is sometimes a little hard to see,” says Rabb. “But insofar as we can see it, anything on the product side would be fair game for experimentation.”

Games are top-of-mind for Rabb, and that strategy isn’t all that surprising. The NYT boasts 10 million daily players to its games. One million of them subscribe to the Games vertical alone (these subscriptions start at $1/week), which includes crosswords and Wordle. That means the NYT’s fun contingent is the size of The Atlantic‘s entire paying audience.

There’s a simple reason for The Atlantic to pursue games. “True, sticky engagement,” notes Rabb. “[It doesn’t offer] the enrichment quotient that you would get from reading The Atlantic cover to cover . . . but I also think that on a micro level, if you gain even the slightest intellectual enrichment from that five minutes a day, it’s time well spent.”

Beyond the allure of gaming, Labs offers an opportunity to explore revenue sources that are off limits to the editorial property. As Rabb suggests, whereas The Atlantic’s paid subscriber base puts some limitations on advertising, there’s no reason why another partner—on top of OpenAI—couldn’t come in to stamp their name on Labs projects.

“There’s a lot of restrictions and how sponsorship can work on the editorial surfaces, and those, by and large, are not the case with Labs,” he says. Rabb goes so far as to tease the possibility of ecommerce initiatives as another source of revenue (though Thompson clarifies that The Atlantic has no ecommerce projects in the pipeline).

Both Rabb and Thompson view Labs as a safe space to prototype the future of The Atlantic—a place where they can attempt to sidestep the risks around nascent technologies and businesses models in publishing. Thompson says he’s “pretty aggressively” reconsidering everything we know about The Atlantic through Labs, “but it’s not like there’s some roadmap, and at the end of nine months, I’ve mapped how The Atlantic will look,” he says. “But I 100% want to study Labs and see if there’s stuff we learn there.”