Getting Started with AI Agents: A Setup Guide for Business Professionals

You do not need a technical background to use an AI agent. You need a paid subscription, a desktop application, and twenty minutes. This guide covers exactly what to install, how to configure it, and how to run your first task - written for professionals who bill by the hour and do not want to read twelve pages of preamble.

If you are new to AI agents and want to understand what they are and why they matter before getting started, see our guide for business leaders.

Key Takeaways

• Claude Cowork is currently the most mature general-purpose AI agent available to non-technical professionals, available on macOS and Windows
• Setup requires a Claude Pro subscription ($20/month minimum), the Claude Desktop application, and a dedicated workspace folder
• Cowork runs in an isolated virtual machine - it can only access folders you explicitly grant permission to
• Your first task should be simple and low-stakes: organize a folder, summarize a set of documents, or create a formatted report from raw notes
• Microsoft's Copilot Cowork, powered by Claude, is entering private preview in March 2026 for Microsoft 365 enterprise customers

The Current Landscape

As of March 2026, the AI agent market has a clear leader for non-technical users: Claude Cowork from Anthropic. Google, OpenAI, and Microsoft are all developing competing products, but Cowork is the one you can install and use today without writing a single line of code.

OpenAI's Operator handles browser-based tasks but is limited to web automation. Google's agent offerings remain developer-focused. Microsoft announced Copilot Cowork - built on Claude's technology - in March 2026, but it is currently in private preview for select enterprise customers through the Frontier program. Broader availability is expected later in the year.

If you are on a Microsoft 365 enterprise plan, Copilot Cowork may become your primary agent interface. If you want to start now, Claude Cowork is the path.

What You Need

A Mac or Windows computer. Cowork is available on both platforms with full feature parity - file access, multi-step tasks, plugins, and all connectors work identically.

A paid Claude subscription. Claude Pro at $20 per month gives you access. Claude Max at $100 or $200 per month provides higher usage limits for heavier workloads. Team and Enterprise plans also include access.

An internet connection. Cowork communicates with Anthropic's servers throughout your session. Offline use is not currently possible.

Installation: Five Steps

Step 1 - Download Claude Desktop. Go to claude.ai/download. Select your operating system and download the installer.

Step 2 - Install the application. On macOS, open the .dmg file and drag the Claude icon into your Applications folder. On Windows, run the .exe installer.

Step 3 - Launch and sign in. Open Claude from your Applications folder or Start menu. Sign in with your Claude account credentials. If you do not have an account, create one during this step and subscribe to a Pro plan or above.

Step 4 - Verify Cowork access. Once signed in, look for the mode selector at the top of the interface. You should see tabs for Chat and Cowork. If you see the Cowork tab, your setup is complete. If not, confirm your subscription is active - syncing can take a few minutes.

Step 5 - Create a workspace folder. Before your first task, create a dedicated folder on your computer - something like "Cowork Projects" in your Documents directory. This becomes the sandbox where Cowork reads and writes files. You can create subfolders for different types of work: Research, Documents, Inbox, Output.

How Cowork Operates

Understanding the architecture matters because it affects what you can and cannot ask the agent to do.

Cowork does not run directly on your computer. It boots an isolated virtual machine - a sandboxed environment that can only access the folders you explicitly share. When you grant Cowork access to a folder, it mounts that folder into the virtual machine. Claude can read and modify files within that mounted folder but cannot touch anything else on your system. Your personal files, browser history, system settings, and other applications are completely isolated.

When you assign a task, the process follows a consistent pattern. You describe your goal in plain language. Claude creates a plan and shows you the proposed steps. You review and approve the approach. Claude executes the task autonomously - this can take minutes or hours depending on complexity. When finished, Claude presents the results for your review.

Throughout execution, Claude surfaces its reasoning so you can follow along. You can intervene to course-correct or provide additional direction at any point. For complex tasks, Claude may coordinate multiple sub-agents working in parallel.

Two important limitations to know upfront. First, Cowork has no memory between sessions - each new task starts fresh. Second, the Claude Desktop application must remain open while Cowork is working. Close the app and the session ends.

Your First Task

Start simple. The goal is to verify your setup works and build intuition for how to direct the agent, not to tackle your most complex project on day one.

Option A - Folder organization. If you have a cluttered Downloads folder or a project directory that has accumulated disorganized files, point Cowork at it and ask it to sort files by type, apply consistent naming conventions, and create a logical folder structure. This is a low-risk task with immediately visible results.

Option B - Document summarization. Place a set of meeting notes, reports, or research documents in your Cowork workspace folder. Ask Cowork to read everything and produce a consolidated summary with key themes and action items. This tests Cowork's ability to read multiple files and synthesize information.

Option C - Report creation. If you have raw data in a spreadsheet or a collection of notes, ask Cowork to produce a formatted Word document or presentation from that material. This tests Cowork's document creation capabilities.

A practical prompt for Option B might read: "Read all documents in this folder. Create a summary document that identifies the three most important themes across all files, lists any action items mentioned, and flags any conflicting information between documents. Save the output as summary.md."

If the task completes successfully and the output is reasonable, your setup is confirmed.

Making Cowork More Powerful

Once you are comfortable with basic tasks, three features extend Cowork's capabilities significantly.

Connectors link Cowork to external services. Gmail and Google Calendar connectors are available, allowing the agent to read your email, draft responses, and coordinate scheduling. Google Drive connectivity lets Cowork work with cloud-hosted documents. DocuSign and FactSet connectors were added in February 2026 for enterprise users. Configure connectors through Claude's Settings panel.

Claude in Chrome pairs with Cowork to give the agent browser access. When installed, Cowork can navigate websites, extract information from web pages, and complete tasks that require internet research - without you manually copying and pasting content.

Skills improve Cowork's output quality for specific file types. Built-in skills for .docx, .pptx, .xlsx, and .pdf handle formatting, layout, and structure more reliably than a generic prompt would achieve. These activate automatically when relevant.

Global and folder instructions let you set standing preferences. You can tell Cowork your preferred tone, format conventions, or role context once, and it applies across every session. Folder-specific instructions activate whenever you are working in a particular directory - useful for projects with consistent requirements.

How to Think About Directing an Agent

The mental model that works best is not "prompting" - it is delegation. The same skills that make someone effective at delegating to a junior colleague make them effective at directing an agent.

Be specific about the end state. What does the finished product look like? What format should it take? What should it include and exclude?

State constraints explicitly. If there are files that should not be modified, boundaries on scope, or formatting requirements, say so upfront. An agent will not always stop to ask for clarification - it will make assumptions and proceed.

Review strategically, not exhaustively. Checking every micro-decision the agent makes defeats the purpose. Check the output against your intent. If the direction is wrong, course-correct. If the details need polish, edit the finished product. The value is in the 80% of mechanical work you did not have to do yourself.

For concrete examples of how to put these principles into practice across seven high-impact use cases, see Seven Ways Business Leaders Are Using AI Agents Today.

What to Be Aware Of

Cowork is a research preview with unique risks due to its agentic nature and internet access. Anthropic's own documentation states clearly: do not use Cowork for regulated workloads. Cowork activity is not captured in audit logs, compliance APIs, or data exports. Conversation history is stored locally on your computer, not subject to Anthropic's standard data retention.

For any work involving sensitive client data, proprietary information, or regulatory obligations, understand the data handling implications before proceeding. Our analysis of AI agent security risks and the security-first deployment framework covers what organizations need to consider. For a broader perspective on enterprise AI data governance, see our insight on how AI data governance mirrors challenges enterprises already solved.