Every releases podcast on human role in AI workflows; introduces Claudie, an AI agent managing client work at consulting firm

Every released a new episode of podcast 'AI & I' featuring Dan Shipper and Kieran Klaassen (GM of Cora), discussing where humans fit in AI-driven work. The framework breaks engineering into four steps—plan, work, review, and compound—with AI handling execution while humans frame problems and judge output quality.

Claudie, an AI agent running on a Mac Mini with a Claude Max account, manages client updates, the sales pipeline, and slide deck creation for Every's consulting team. Instead of building separate agents for each function, capabilities were converted into plugins within Claudie after it consistently exceeded expectations.

Claudie's permissions are tied to a 'trust battery' (a concept coined by Shopify CEO Tobi Lütke). A judge agent reviews her work nightly and adjusts her battery level by up to ±5 percent based on performance; Claudie reads the verdict each night and updates her memory and behavior to avoid repeating mistakes. Her battery started at 20 percent, compared with a human new hire's typical starting point of about 50 percent.