Will The Washington Post Embrace the AI Slush Pile?

Yahoo

07-06-2025

The Atlantic Daily, a newsletter that guides you through the biggest stories of the day, helps you discover new ideas, and recommends the best in culture. Sign up for it here.
Early in my career, I worked as an assistant at a literary agency. Big publishers generally consider taking on only writers already represented by agents, which makes literary agencies a front line of sorts. As the person opening the mail, I was the front line of the front line. I saw the true democratic range of the slush pile, full of pitches that no one had vouched for and, for the most part, that no one ever would.
One thing I learned: There's a lot of writing out there that you, the reader, just don't need in your life. Some of it is inaccurate. Some of it is self-serving. Above all, a lot of it is just not interesting enough to find many readers: generic, predictable, telling you something you've already heard. Editorial gatekeepers get a bad name, but from another point of view, they are heroically holding back a tidal wave of crap while, ideally, letting the good stuff through.
That early lesson in the value of editorial judgment came back to me this week, when The New York Times reported on an effort taking shape at The Washington Post under its owner, Jeff Bezos, and its publisher, Will Lewis. Through a project internally called Ripple, Post executives intend to dramatically expand opinion writing at the paper, creating an offering outside the paywall that will include content from partner news organizations and Substack. More controversially, a final phase of the plan will employ an AI writing coach called Ember to assist 'nonprofessional writers' in submitting op-eds.
This effort was not exactly news to me. Until January, I was a senior editor at the now almost completely hollowed-out Opinions section of the Post. Along with others from across the organization, I'd participated about a year ago in a brainstorming session on what would become Ripple. At that point, it was clear already that Bezos was interested in massively scaling up the output of our section, perhaps on the model of Amazon—which had scaled up and up for years before turning a profit. It had also become clear that the way to management's heart was to cite artificial intelligence as the means to any end, a special technological sauce to be drizzled on everything.
[Damon Beres and Charlie Warzel: At least two newspapers syndicated AI garbage]
Although I'd started out skeptical, by the end of the session, I was convinced that the Post did have the potential to reach a larger audience. Readers want locally relevant news, but local outlets are succumbing one by one to the dynamics of a centralized online market for both content and advertising. If we were so determined to scale up, why couldn't the Post partner with existing local news sources, offering them a tech back end, a network effect, and a cut of resources while tapping a much larger pool of locally written and edited work?
Some elements of those ideas seem to have trickled into the Ripple project. But another vision was presented at the brainstorm too, the spark of what is now called Ember. The concept seemed to be that anyone could write a good op-ed, if only they had coaching from an AI editor. As a newsroom AI strategist explained the premise to the members of the group—most of whom had never faced an inbox of op-ed submissions—I felt filled with dread, because the content that this program would yield sounded dreadful.
When you consider pitches, as an editor of opinion content, you look for surprise: insightful analysis of new information, diagnosis of and perhaps solutions to an unappreciated problem, a personal tale told in a way that makes you laugh or tear up, an original way of experiencing something familiar. Generative AI based on large language models, by contrast, is optimized to produce writing with the opposite qualities. It is a predictability machine, operating by asking what word is most likely to come next in a sentence based on all the other text that has fed into its training data. This doesn't mean that AI can't be a useful tool for certain kinds of writing and editing, but it does mean that an AI editor will probably exacerbate exactly the qualities that make the opinion slush pile so slushy in the first place. What Ember seems likely to produce, in other words, is the kind of writing I have spent my whole career trying to hold back.
There are good uses for AI at a newspaper, which is why it's so puzzling that everyone keeps trying to make AI do not those tasks but the ones it is bad at, the ones that we humans most want to keep for ourselves. Just to take one example, much of the Washington Post archive is inaccessible via search; why not use AI to crawl, tag, and make discoverable this huge body of work?
Maybe because that's not showy enough. Everyone these days seems to want to make AI the writer, the editor, the creator, the star. But within media, at least, that's not the best use case for this technology. Instead, we should be using these powerful tools for scut work, data crunching, even brainstorming—not as a substitute for the editorial judgment and critical thought that make writing worth reading.
'The values of The Post do not need changing,' Bezos wrote when he bought the newspaper, in 2013. 'The paper's duty will remain to its readers and not to the private interests of its owners.' That declaration became harder to trust last year, when Bezos blocked the Post editorial board's endorsement of Kamala Harris, on the pretext that he had suddenly decided that making presidential endorsements gave the impression of bias—only to cheer the 'extraordinary political comeback and decisive victory' of Donald Trump just a few weeks later.
[Listen: The media is splitting in two]
The proposed use of Ember casts doubt on the commitment to readers too. If Bezos is really interested in scaling up opinion writing beyond the scope of human editors, what he's essentially doing is either starting a social blogging network—something like Medium, Reddit, or Substack—or a contributor platform of glorified press releases, such as the one abandoned in 2018 by HuffPost as a drag on its brand. As the New York Times columnist Lydia Polgreen wrote on X, 'When I was editor of HuffPost we shut down our contributor platform because it was bad for our journalism and it did not contribute significant traffic or revenue.' If Bezos wants to run a social network, perhaps to compete with Elon Musk for clout, that's fine. But let's not pretend that it's journalism, or that it's good for the Post and its readers.
As recently as 2021, when the paper had already been under Bezos's ownership for years, the Post was touting a plan to add 41 new editors to the newsroom. 'This expansion demonstrates anew that The Washington Post is an ascendant news organization, with boundless ambitions and a growing capacity to meet them,' wrote then–Executive Editor Sally Buzbee and her team.
That memo might as well be from another universe. Today, the Post's owner seems to have lost track of those ambitions, or replaced them with other ones. But if the journalistic or commercial health of the paper as an institution still matters to him, I hope he will realize that using AI to scale up the slush pile is a poor idea. If that's really Bezos's dream for the nation's readers, he should pursue it separate from the Post, rather than risk undermining the editorial tradition that has made the paper great.
Article originally published at The Atlantic