Every's 30 people, four products and one cloned editor: the self-driving company in practice

Dan Shipper told Platformer's Casey Newton that AI now writes essentially all of the code at Every, the roughly 30-person company he co-founded in 2020 with Nathan Baschez, while humans still mostly write the essays ^[1][2][3]. That matters because Every is one of the few firms running the "self-driving company" idea in public, the phrase Replit's Amjad Masad used for an operation where most engineers no longer look at the code ^[4].

The shape is unusual. Every publishes on AI, including Shipper's Chain of Thought column, his AI & I podcast, and the "vibe checks" in which he and colleagues get early access to frontier models before general release ^[5]. It also ships four products: Cora, an email assistant; Sparkle, a file organizer; Spiral, a writing tool; and Monologue, a dictation app, all bundled with the journalism into a $20-a-month subscription ^[6]. That is $240 a year per subscriber carrying both a newsroom and four software lines ^[7].

Then the number that complicates the pitch: Every doubled from about 15 people to around 30 over the past year while loudly automating everything it could ^[8][9]. Shipper's explanation is that AI is "trained on the residue of human expertise" and cannot see past it ^[10]. That is the load-bearing assumption of the whole model, namely that automation lowers the cost of execution and the savings get spent on people who can produce expertise no training set contains yet. It is an assertion rather than a demonstration, and the accounting that would settle it is not public.

The clone is the sharper experiment. According to Shipper, Every collected a dataset of 30,000 historical edits by editor in chief Kate Lee, used it to build a copy-editing agent, and back-tested that agent against her past work ^[11]. Newton frames this as capturing the expertise of a single employee and distributing it across the enterprise, and as a preview of how more businesses will treat the relationship between AI and staff ^[12]. Operators should also read it as an unsettled compensation question. The taste is the asset, the dataset is a copy of the asset, and an agent fitted to 30,000 past edits is by construction a good model of last year's standard rather than next year's.

The second strain is independence. Every reviews the models of labs whose models it also builds on ^[13], and it published a critical review of Sonnet 5 under the line "a model pitched for everyone impresses no one" ^[14]. Shipper says the labs ask what Every thinks even before publication, because they want to improve the model and would rather hear it early than from a crowd of users ^[15]. They would probably prefer no public teardown, he says, but know the intent is not to be mean ^[16]. His defence of the arrangement is that the arbiter role may be one of Every's most durable assets: "No one trusts a model company to tell you where they objectively sit" ^[17]. That is an asset and a supplier dependency occupying the same seat. Newton, for his part, disclosed that his fiance works at Anthropic, whose models Every reviews and builds on ^[18].

One more Shipper claim is unverifiable and will be quoted at editorial leaders anyway: that almost every writer is now using AI, and most are not saying so ^[19].

Watch whether Every's headcount keeps climbing at the same rate once the easy automation is done, whether Lee's job description changes as the copy-editing agent matures, and whether a negative review ever costs Every its pre-release access.