Claude Cowork for Architecture and Engineering Firms: What It Is and How to Start Safely

Most architects use AI as a chat: you ask a question, you get an answer. Claude Cowork works differently. You describe the finished result you want, and Claude plans and carries out the steps to get there, working with your files and connected apps.

That makes it useful for the tedious middle of office work. It also means you need to set it up carefully. This guide covers both.

Everything below about how Cowork works comes from Anthropic's own documentation, checked in October 2026. Features change often, so check the official Cowork guide for the latest details.

What Claude Cowork is

According to Anthropic, Cowork brings the agentic approach of Claude Code to everyday knowledge work, without needing a terminal. Instead of a single reply, Claude takes on multi-step tasks and carries them out for you. Sessions run in the cloud, so a task can keep going after you close your laptop.

Who can use it: Cowork is for paid Claude plans only (Pro, Max, Team, and Enterprise). It works in the Claude desktop app on Mac and Windows on all paid plans. Web and mobile access depends on your plan, and on Enterprise an admin has to enable it.

What it can access:

• Your files: on desktop, Claude can read and write files only in the folders you connect.
• Your apps: through connectors you choose to turn on.
• A browser: on desktop, it can open websites, read pages, click, and fill in forms.

Good first tasks for architecture and engineering firms

Anthropic suggests starting with a task you already know well, so you can judge the result quickly. Good candidates have several inputs, produce a file, repeat often, and have a boring middle.

Here are ideas that fit that pattern. They are suggestions to test, not reported results:

• Meeting minutes from notes: give Cowork a folder of rough notes from one project, and ask for formatted minutes with decisions, open questions, and action items by responsible party.
• A project status summary: point it at a project folder and ask for a one-page status summary for the client, in plain language.
• Consultant comment tracking: give it several consultant comment letters, and ask for one table of comments, responsible party, and status.
• Product research table: give it manufacturer data sheets, and ask for a comparison table. Tell it to write "not found" when a data point is missing.
• A weekly report: combine the same few sources into the same report every week.

Tasks to keep away from Cowork: code compliance conclusions, life safety decisions, sealed documents, and anything you would not be comfortable reviewing line by line before it goes out.

Set it up safely

Cowork can act, not just answer, so set it up carefully.

Connect only the folders you need. Claude can only read and write in folders you've connected. Start with a copy of one project folder, not your whole server.

Choose the right approval mode. Anthropic offers three:

• Manual: Claude asks before it acts. Start here.
• Auto: Claude reviews each action for safety and blocks unsafe ones. It uses more of your usage limit.
• Skip: Claude acts without asking. Anthropic says to use this only when you completely trust every action, connector, file, and app involved.

Review the plan first. Anthropic recommends reviewing Claude's planned actions before letting it continue, especially with sensitive files. Claude asks for your explicit permission before permanently deleting files.

Remember the risks. In Anthropic's words, "Cowork has unique risks due to its agentic nature and internet access." Be careful which websites and third-party connectors you trust.

Check your firm's rules on client data. If you handle confidential client or government projects, check your contracts and your firm's AI policy before you connect any project files. Team and Enterprise plans give admins more control.

Your first session

1. Open the Claude desktop app and start a Cowork task.
2. Connect one folder with real, non-sensitive material, such as an old, completed project.
3. Describe the finished deliverable you want. For example: "Create a one-page summary of this project's key decisions for a new team member."
4. Ask Claude to check its understanding first. Anthropic suggests: "Before we begin, repeat my ask back to me so we're aligned, then ask me as many clarifying questions as you have."
5. Review the result line by line, just as you would a junior colleague's first draft.

The same rule still applies

Cowork does more of the work, but the rule doesn't change: AI prepares, professional judgment decides. Review every output before it reaches a client, a consultant, or a permit set.

Learn the basics first

If you're new to AI, start with plain chat before Cowork. Our guide on how to use AI in architecture and the free cheat sheet of 10 AI prompts for architects are good first steps. The book AI for Architects covers everyday AI habits, verification, and a simple office AI policy.