The AI paradox: More automation, more humans, more work | Dan Shipper

Dan Shipper argues that the near-term AI story is not mass replacement but a paradox: models make yesterday's competence cheap while creating more work for humans who must direct, review, and extend those capabilities. His evidence is Every, a roughly 30-person company whose engineers, designers, writers, editors, salespeople, and support staff all experiment with new models. He treats the company's practices as an early view of the future rather than a forecast detached from experience. He predicts two linked work surfaces: company-level agents that people delegate work to, often through Slack, and computer-based environments such as Codex or Claude Cowork that become an operating system for email, documents, research, and other knowledge work. The discussion moves from the initial idea of one personal agent per employee toward a company super-agent because agents currently need a human owner who maintains context, fixes failures, and makes the system useful. Personal agents may become more practical as models improve, but the immediate pattern is a general agent at the top that later develops specialized team agents. The other major shift is that SaaS tools may run inside the agent environment rather than embedding a separate AI surface. This lets users bring their own model tokens and lets products support humans and agents simultaneously, potentially increasing SaaS usage rather than destroying it.