Cowork

Overview

Cowork is a research-preview tool from Claude/Anthropic that gives the assistant access to files inside a designated folder so it can read, edit, and transform source materials into finished outputs. It is positioned around automating non-technical knowledge work such as drafting documents from scattered notes, generating spreadsheets from screenshots or raw files, and executing multi-step tasks across local content with less manual copying and coordination.

For AI Product Managers, Cowork matters because it points to a practical interface pattern for agentic work: scoped file access plus multi-step execution. Rather than acting only inside a chat box, the assistant can operate on real working materials, making it useful for synthesis, reporting, document creation, and operational workflows. At the same time, newsletter coverage highlights the tradeoffs of this model—especially workflow friction when users must repeatedly grant folder access, copy files, and define plans—making Cowork a useful case study in usability, trust, and context management for AI-powered productivity tools.

Key Developments

• 2026-01-13 — Claude introduced Cowork as a research preview that lets the assistant access, read, and edit files in a designated folder to automate non-technical tasks like drafting documents from notes or generating spreadsheets from screenshots.
• 2026-01-18 — Clement Delangue highlighted local model support in Cowork, emphasizing an on-device option for users who want to keep data local instead of sending it to the cloud.
• 2026-02-11 — Cowork launched on Windows, reaching feature parity with macOS, including file access, multi-step task execution, plugins, and MCP connectors. This update was also noted by Boris Cherny.
• 2026-03-03 — Santiago criticized Cowork’s multi-step file workflow, arguing that granting folder access, copying files, and generating a plan creates friction when handling many files or frequently changing inputs, and suggested native chat-based file support as a smoother alternative.

Relevance to AI PMs

1. Evaluate agent UX beyond chat. Cowork is a concrete example of how assistants can move from prompt-response interactions into file-based workflows. AI PMs can study where scoped file access improves task completion versus where setup friction reduces adoption.

2. Design for trust, privacy, and permissioning. Because Cowork works on local folders and later added support for local models, it surfaces important product questions around data boundaries, enterprise privacy, and how much autonomy users are willing to grant an assistant.

3. Prototype operational automations. PMs can use tools like Cowork to accelerate recurring non-technical work such as PRD drafting, spreadsheet generation, synthesis of customer notes, and packaging research artifacts from messy source materials.

Related

• Claude — Cowork is introduced as a Claude research preview and reflects Claude’s broader push into agentic desktop/file workflows.
• Anthropic — The company behind Claude; Cowork is part of its product direction around practical AI assistance.
• Claude Desktop — Closely related as the desktop environment where Claude-based local workflows and integrations may be experienced.
• MCP — Mentioned in connection with Cowork’s connectors; MCP expands how tools and data sources can be integrated into assistant workflows.
• desktop-commander-mcp — Relevant as an example of the MCP ecosystem around desktop and file-oriented assistant actions.
• Boris Cherny — Covered Cowork’s Windows launch and adjacent Claude workflow updates.
• Santiago — Offered critical feedback on Cowork’s file workflow friction, highlighting UX tradeoffs.
• Clement Delangue — Highlighted Cowork’s local model support, reinforcing the privacy/on-device angle.