Foreword: The evolution of AI at work

You lead a team that doesn't write code. The work is contracts, decks, financial models, pipeline reviews, regulatory filings, executive reports. AI has been on every board agenda for two years and while engineers at your company are power users, most of your organization still uses it like a chatbot: ask a question, get an answer, go do the work yourself.

The employees seeing real results are the ones that embed AI into how work actually gets done. That means using AI that reads the same spreadsheets your analysts read, drafting against the same templates your legal team uses, and updating the same CRM your reps live in, then handing back finished work rather than suggestions.

The trajectory has been fast. In 2024, using AI at work meant interfacing with a chat window. In 2025, Claude Code (opens in new tab) put an agent at the command line and developers started delegating hours of work at a time. In 2026, Claude Cowork (opens in new tab) brings that same capability to the desktop for everyone else at your company: the analysts, lawyers, account executives, product managers, marketers, and more. And with plugins (opens in new tab), skills (opens in new tab), and commands (opens in new tab), Claude becomes even more customizable and capable.

This guide walks through how to deploy Claude Cowork (opens in new tab) across a business function: where to start, how to structure a pilot, and best practices for scaling what works. The examples come from Anthropic's own finance, legal, sales, and product teams, alongside customers running Claude Cowork in production, including Thomson Reuters, Zapier, and Jamf.

While the use cases are specific to Anthropic and other enterprises, the best practices and patterns are not. We hope you find these lessons learned and best practices applicable to your own workflows.