The Coworker Bet

Anthropic pushed Claude Cowork to web and mobile this month. The agent drafts emails in Outlook, puts meetings on the calendar, edits SharePoint files. It does not wait for a prompt. It sits in the workspace and carries part of the workload.

This is not a feature update. It is a product thesis. The chat window model — type a question, get an answer, copy it somewhere — tops out. Every team I have worked with that uses AI this way sees the same curve. Fast initial gains, then a plateau around month three. The tool helps, but it never compounds because it never touches the actual work.

The coworker model is different. The agent has context from your documents. It drafts the follow-up you were going to write. It handles the logistics so the human handles the judgment. Two entities splitting the work by what each does best.

Microsoft surveyed 20,000 workers and found only 16% operate this way. The rest prompt. Both groups have the same models. The gap is not capability. It is whether you redesigned the workflow or just added a chat tab.

Anthropic is betting the coworker model wins. Given what I have seen in every deployment this year, I think they are right.