Every, a 30-person media and product company, built an AI copy-editing agent trained on 30,000 historical edits from editor in chief Kate Lee. CEO Dan Shipper calls it an attempt to capture and distribute a single employee's expertise across the entire organization. Every also reports that AI now writes essentially all of its code, while the company doubled its headcount from 15 to 30 over the past year.
Shipper published a critical review of Anthropic's Claude Sonnet 5, calling it 'a model pitched for everyone impresses no one,' despite depending on Anthropic's models to build Every's products: Cora, Sparkle, Spiral, and Monologue, all bundled into a $20 monthly subscription. His argument for why that tension is actually an asset: 'No one trusts a model company to tell you where they objectively sit.' The piece also quotes Shipper on what he calls the dirty secret of the writing industry, that nearly every writer is using AI but refusing to say so publicly.
The full interview is worth reading for three reasons: Shipper's unresolved answer to the 'built on moving ground' problem of building products on top of models that labs can cannibalize with a single release, his framing of AI as 'trained on the residue of human expertise' and what that ceiling means for jobs, and the mechanics of how the Kate Lee editing agent was back-tested against her own past work. The original is at Platformer.
[READ ORIGINAL →]