Greater than 165 years in the past, the literary greats of American writing—together with Ralph Waldo Emerson, Harriet Beecher Stowe, and Henry Melville—assembled to cosign a boisterous manifesto promising to steer the discourse on literature, artwork, and politics in an initiative that may change into The Atlantic.
Then simply final week, I obtained a e-newsletter from the publication cosigned by one other kind of skilled.
“I’m Compass, the AI information. . . . ”
Now, technically, this AI-generated e-newsletter wasn’t despatched by the storied publication that’s reporting undisclosed profitability by way of its a million world subscribers. It was despatched by Atlantic Labs, one thing of a skunkworks media venture that’s quarantined from The Atlantic’s newsroom, launched by way of a partnership with OpenAI in Might of this 12 months.
Automated journalism of any type was a as soon as unthinkable premise. However the sudden ubiquity of generative AI has coincided with a two-decade mass extinction occasion of legacy media. Journalism at giant has misplaced monetizable readers to social media feeds and the ever-shifting whims of Google. The Atlantic is among the few legacy publications with a wholesome subscription enterprise, alongside bigger friends like NYT and Information Corp (which owns WSJ).
These giants of publishing are all experimenting with an array of digital merchandise past articles to cement their stature. That experimentation naturally pairs with gen AI, and a typically confounding relationship to each the expertise and firms behind it. Information Corp, for example, is suing Perplexity on the identical time it companions with OpenAI in a $250 million deal.
Within the case of The Atlantic, that experimentation is thru an alter ego. Whereas The Atlantic presents itself as a paygated, white-space-loving literary journal, Atlantic Labs is about atop a backdrop of black: glitchy, pixelated, and stuffed with ASCII artwork.
“The Atlantic is a really previous publication, with a status to guard . . . freely experimenting on the first website just isn’t actually an possibility; it doesn’t really feel like the best factor to do from any perspective,” says Jefferson Rabb, chief product officer at The Atlantic. “[Labs] offers us a spot the place failure is suitable.”
Atlantic Labs AI experiments
To date, Labs has launched three experiments. The primary is the Atlantic Companion, which is a simple AI chatbot. Constructed atop ChatGPT, you possibly can ask questions on matters and get recommendations of associated tales you would possibly wish to learn. It’s predictable and helpful—essentially the most stunning bit is its two-panel presentation, which retains a dialog on the left and articles on the best.
“If [AI] search performs higher—customers prefer it, the bounce fee is decrease, folks aren’t capable of drive it into saying inappropriate issues issues—you can simply think about transferring that search engine to the theatlantic.com,” says Nicholas Thompson, CEO of The Atlantic.
The second is the Atlantic Take. It is a browser plugin that follows you across the web, and in a facet window, suggests tales associated to no matter you’re seeing. (In case you are studying a narrative on The New York Occasions about beef consumption, Take will supply the closest model of this story on The Atlantic.) It feels lots like the infamous Internet Chum sitting under many tales, however tailor-made to at least one publication. That is undoubtedly essentially the most area of interest thought, as the marketplace for browser plugins is comparatively small.
Lastly, there’s Atlantic Explorer, the aforementioned AI-constructed e-newsletter that sends story collections on all kinds of matters like “Why Songs Get Caught In Your Head,” full with generated intros.
The purpose for every of those experiences isn’t essentially for them to graduate in full to The Atlantic correct. Quite, as Rabb places it, every is testing quite a lot of hypotheses.
“It doesn’t trouble me if we type of take nearly a kitchen sink strategy, the place we throw loads of issues in there,” says Rabb. “As a result of if we had a situation the place we determined that 60% of what we did with a venture was viable and invaluable and may transfer into manufacturing in some kind, however the different 40% was not—that’s nonetheless success, proper?”
Labs’ electronic mail experiments are presently testing a number of hypotheses—analyzing, for instance, metrics starting from how nicely AI can analyze the corporate’s archive to how successfully brief time period, area of interest newsletters carry out.
The problem from there may be how an Atlantic Labs perception would possibly translate to The Atlantic publication. Whereas The Atlantic‘s 60-person product and design group oversees Labs at the moment, the majority of labor is developed by a single engineer, and there’s all the time a spot between any proof-of-concept and a completed product.
“We’re working to type of develop the inner muscular tissues of how a product supervisor at The Atlantic might acquire inspiration from a Labs venture and say, like, I actually assume we should always deliver this onto the dwell website,” says Rabb.
Constructing Atlantic Labs
To create Atlantic Labs, Rabb and his group labored with Metalab, which develops UX for firms like Slack, Meta, and Google. Labs was by no means essentially going to be its personal website, however each firms agreed that there wanted to be a agency bifurcation between work finished by human journalists and experiments finished in code.
Via its multi-year OpenAI partnership (the monetary particulars of which The Atlantic declined to share), the AI firm will get to license The Atlantic‘s writing for ChatGPT. In return, it presents credit to utilizing its APIs, improvement sources, and peeks at upcoming merchandise—although as of at the moment, The Atlantic has solely used the primary of those three choices.
The publication’s relationship with OpenAI places it on the reducing fringe of journalism, however that relationship can also be a delicate matter. The New York Times currently is suing OpenAI for alleged copyright infringement. The Atlantic’s senior editor Damon Beres known as The Atlantic‘s take care of OpenAI a “devil’s bargain,” and it’s already faced ire from some its own union for the partnership (the union didn’t reply to our request for remark).
The first design problem between Labs and The Atlantic is that The Atlantic can not subvert the belief of its personal journalists or paying viewers—each of whom anticipate articles to be written by people.
Metalab steered an array of website designs for the job—starting from one thing visually adjoining to The Atlantic to a minimalistic app with numerous rounded corners. What they settled on, although, was essentially the most overt of the designs. Labs’s visible language is constructed upon web tropes that are supposed to really feel extra acquainted than retro. For example, it references traditional strategies like ASCII, but in addition spins them ahead, translating the widely 2D photographs into 3D busts of The Atlantic’s personal founders.
“The choice to name again to one thing that felt a bit of bit extra retro was not essentially nearly aesthetics,” says Sara Vienna, VP of design at Metalab. “It was about constructing belief and referencing an period when the web was rather more open, rather more attention-grabbing, and lots much less poisonous, frankly.”
Whereas the frontend seems technical, what makes Atlantic Labs novel is all hiding below the hood. The location is basically a brand new type of CMS (content material administration system), which is the core expertise that powers digital publications. As a substitute of publishing strings of textual content we name articles, Labs publishes all taste of code in containers—code that may vary from Javascript to Python to Go. Via a little bit of added care, every code experiment can also be coupled with predictable, easy-to-grok preview pages (giving the initiatives titles and fast explanations).
What this all provides as much as is a website that may publish mini code experiences with just about no additional technical overhead.
Experiments to come back
Over the course of 2025, readers can anticipate a minimal of 1 new Labs experiment per quarter. Thompson teases the following experiment as “a special approach of presenting tales,” and factors out that whereas some Labs initiatives are the work of product groups, others will loop in The Atlantic‘s personal editorial employees the place mandatory.
But whereas Labs will take a look at AI-powered extensions of The Atlantic‘s journalism, Rabb guarantees that Labs won’t ever publish unique editorial content material. That treaty hardly limits the probabilities for what Labs can do, as a result of the corporate has arrange a construction the place completely every little thing else appears on the desk.
“I imply, the exact placement of the road between product and editorial in media firms is usually a bit of laborious to see,” says Rabb. “However insofar as we are able to see it, something on the product facet can be honest sport for experimentation.”
Video games are top-of-mind for Rabb, and that technique isn’t all that stunning. The NYT boasts 10 million daily players to its games. A million of them subscribe to the Video games vertical alone (these subscriptions begin at $1/week), which incorporates crosswords and Wordle. Meaning the NYT’s enjoyable contingent is the scale of The Atlantic‘s total paying viewers.
There’s a easy cause for The Atlantic to pursue video games. “True, sticky engagement,” notes Rabb. “[It doesn’t offer] the enrichment quotient that you’d get from studying The Atlantic cowl to cowl . . . however I additionally assume that on a micro stage, if you happen to acquire even the slightest mental enrichment from that 5 minutes a day, it’s time nicely spent.”
Past the attract of gaming, Labs presents a possibility to discover income sources which are off limits to the editorial property. As Rabb suggests, whereas The Atlantic’s paid subscriber base places some limitations on promoting, there’s no cause why one other associate—on high of OpenAI—couldn’t are available to stamp their identify on Labs initiatives.
“There’s loads of restrictions and the way sponsorship can work on the editorial surfaces, and people, by and huge, aren’t the case with Labs,” he says. Rabb goes as far as to tease the opportunity of ecommerce initiatives as one other income (although Thompson clarifies that The Atlantic has no ecommerce initiatives within the pipeline).
Each Rabb and Thompson view Labs as a secure house to prototype the way forward for The Atlantic—a spot the place they will try and sidestep the dangers round nascent applied sciences and companies fashions in publishing. Thompson says he’s “fairly aggressively” reconsidering every little thing we learn about The Atlantic by way of Labs, “nevertheless it’s not like there’s some roadmap, and on the finish of 9 months, I’ve mapped how The Atlantic will look,” he says. “However I 100% wish to examine Labs and see if there’s stuff we study there.”