I am finalizing the intro for this newsletter while up in the air on my way to Asia. It has been a trip that has been planned for a long time, got cancelled twice, got shortened because I only have so many vacation days now, and even at the airport on Sunday morning, it almost didn't happen. A transit visa to China, that allows Americans to stay in the country for 10 days without the need to get a full visa was not being accepted at the checkin counter. The rules are very strict, and I won't get into details here, but being a seasoned traveler (did I mention here that I was a travel writer for 10 years?), I have done my research well in advance and had all the details checked, so I was sure that all I did was correct. Took them almost 2 hours to figure it out, so the intro that I planned to write and schedule while in the lounge, didn't happen. I am now finally seating on an almost 18 hour flight to Singapore, before connecting to Shanghai in China.
You must have gasped at that thought… at being confined to an airplane that long. I did it, so I decided that the "right” way to do this flight is in business class, but of course, this being a personal trip, I didn't want to pay for the price of a business class trip. So what did I do? Almost two years ago I got ChatGPT to figure out how I could use all my miles and points from my credit cards to get it. I had a lot of miles and points scattered, but didn't have enough at the time for a business class ticket, so it also made a plan for me to get the remaining points I needed in the next few months. ChatGPT gave me all the instructions on the cheapest way to do it, and the step by step to how to transfer the miles and here I am. After trying to sleep for the first 6 hours of the flight, to get my body acclimated to the new time difference(+13 hours), I am enjoying the free wifi, and doing some personal light work - mainly my newsletter, LinkedIn and editing my book. Now that you have read all my rumblings, time to get to the hottest topic these days(other than the Pentagon-Anthropic-OpenAI saga): Claude Cowork.
….
There’s a pattern I’ve seen across organizations in my research and my daily work: people love AI tools in demos, then struggle to integrate them into real workflows. The gap between “impressive assistant” and “reliable colleague” has always been the sticking point. I use Copilot at work, and has been getting significantly better in that realm and giving access to the newest models including Claude, but I don't have Copilot to manage my other work/ personal life, so I decided to test Claude Cowork.
Cowork is an autonomous desktop agent built into the Claude Desktop app. It’s not a chat interface with a new skin. Think of it this way: chat is asking a colleague questions. Cowork is delegating a project and checking back when it’s done.
When you open the Cowork tab, you give Claude access to a folder on your computer. From there, Claude can read, edit, and create files:Word documents, PowerPoint decks, spreadsheets, PDFs. All of this, without you uploading anything manually, without you reviewing every intermediate step, and without you reformatting outputs afterward.
You describe what you need. Claude plans the work, breaks it into subtasks, executes them (sometimes in parallel using multiple sub-agents), and delivers finished files directly to your folder.
This is what separates Cowork from every other AI tool in the enterprise stack right now.
If you’re in a role that involves synthesizing information into documents, organizing large volumes of files, preparing presentations from scattered notes, or managing recurring knowledge work, this is directly relevant to you. You don’t need to be a developer. Cowork was explicitly built for non-technical users doing non-coding work.
Step 1: Check Your Subscription
Cowork is available on paid Claude plans: Pro ($20/mo), Max ($100–$200/mo), Team ($30/user/mo), and Enterprise. You’ll need one of these to access the feature. Note that Pro users may hit usage limits faster on complex tasks, since Cowork is more compute-intensive than standard chat.
Step 2: Download Claude Desktop
Go to claude.com/download and select the version for your operating system. Cowork is available on both macOS (requires Apple Silicon, M1 or later) and Windows. Once downloaded, open the installer and launch from your Applications folder (Mac) or Start menu (Windows).
Step 3: Sign In and Find the Cowork Tab
Once you’re in Claude Desktop, look for the mode selector at the top. You’ll see three tabs: Chat, Cowork, and Code. Click Cowork to enter Tasks mode.
Step 4: Select a Folder
At the bottom of the Cowork interface, you’ll see “Work in a Folder”. Click it, then select the local folder you want Claude to work in. This is the boundary of Claude’s access, it can only read and write within this folder.
Practical tip: Start with a low-stakes test folder, not your primary project directory. I named it ClaudeCowork so I will always know what I am adding where. Create a copy of real files you want to reorganize or process, and run Cowork there first to understand how it operates.
Step 5: Set Global Instructions (Optional but Recommended)
Before your first task, go to Settings > Cowork > Global Instructions. This is where you tell Claude how you like to work, like your preferred document format, your role context, your tone preferences. These instructions apply to every Cowork session automatically.
You can also set folder-specific instructions that activate only when working in a particular folder. For example: “This folder contains client reports. Always format outputs with an executive summary at the top and use our standard section headers.”
Step 6: Describe Your Task
Type what you need in plain language. Cowork works best with outcome-oriented descriptions rather than step-by-step instructions. Instead of “open this file, then copy column B, then...” try “Analyze this spreadsheet and produce a Word report summarizing spending by category, with an executive summary and a table of the top 5 expenses.”
Claude will show you its plan before executing. You can review, approve, or redirect.
Here's an example of organize and categorize recent screenshots:
Step 7: Monitor and Steer
While Claude works, you’ll see progress indicators for each step. You can watch, intervene, or walk away. If something looks off, jump in and redirect mid-task. For potentially destructive operations like file deletion, Cowork will explicitly ask for your permission before proceeding.
Based on some testing, Cowork handles these categories well:
Document creation from raw inputs. Give it a folder of meeting notes, a data export, and a brief, and it will produce a formatted report or deck. The output is an editable file, not a PDF screenshot, not a chat artifact.
File organization at scale. Sorting, renaming, and restructuring large volumes of files based on content, type, or date, tasks that would take hours manually.
Research synthesis. Feed it sources and a structure, and it produces synthesized documents that follow your formatting preferences, not a generic template.
Scheduled and recurring tasks. Type /schedule in any Cowork session to set up automated tasks. This is a capability that doesn’t exist in standard AI chat, and it opens the door to genuine workflow automation, not just faster one-off tasks. In the last section I have a list of 30 examples to use Claude Cowork today.
This is the question I get most from enterprise teams. Here’s the honest breakdown:
Integration depth. Copilot’s advantage is native integration with Microsoft 365, as it lives inside Word, Excel, Teams, and Outlook. If your workflows are built on M365, Copilot can surface in context and reduce app-switching. Cowork works at the file system level, which is more universal but less embedded in specific applications.
Autonomy. Cowork operates with significantly more agency. It can plan, decompose, execute multi-step tasks, and coordinate parallel sub-agents without constant human input. Copilot is strong at in-document assistance like drafting, summarizing, reformatting, but it typically operates on a single task at a time within a single application. The distance between “suggest” and “complete” is larger with Copilot.
Data privacy. Cowork stores conversation history locally on your computer, not on Anthropic’s servers. Files never leave your machine for training or cloud storage. For teams handling sensitive data, this matters. Copilot operates within Microsoft’s enterprise cloud, which has strong compliance credentials but different privacy architecture.
Access and cost. Copilot for M365 is $30/user/month (on top of M365 licensing). Claude Pro with Cowork starts at $20/month for individuals; Team plans are $30/user/month. The cost comparison depends heavily on what you’re already paying for.
Flexibility. Cowork connects to external tools via MCP connectors, Gmail, GitHub, Slack, Notion, and more. It can also pair with Claude in Chrome for tasks requiring browser access. This creates a more flexible automation surface than Copilot’s Microsoft-ecosystem focus.
Bottom line: If your team lives in Microsoft 365 and needs AI assistance within those tools, Copilot is a strong fit. If you need an AI that can take on complex, multi-step knowledge work across your file system and connected services without constant supervision, Cowork is worth serious evaluation. Please do not download or use Cowork in your work environment if your company didn't approve it.
All three major AI platforms: Anthropic, OpenAI, and Google, have now entered the agentic desktop space. Here’s where they actually differ for knowledge workers.
ChatGPT Agent (OpenAI)
OpenAI’s ChatGPT Agent launched in mid-2025 and represents the most direct competitor to Cowork in terms of positioning. It runs tasks using a virtual computer, a sandboxed remote browser environment, where it can navigate websites, fill out forms, execute code, and create documents. The key difference is where the work happens: ChatGPT Agent operates primarily in the cloud and excels at web-based tasks (booking travel, scraping sites, managing online workflows). Cowork runs locally on your machine, which means it works with your actual file system and processes data that never leaves your computer.
For knowledge workers doing internal work, synthesizing documents, organizing project folders, generating reports from proprietary data, the local architecture matters. You’re not uploading sensitive files to a remote server to process them. For tasks that require navigating the public web or automating online services, ChatGPT Agent has a strong advantage.
One notable limitation: Plus subscribers get 30 to 40 uses per month with ChatGPT Agent, which is restrictive for heavy workflows. Cowork ties into your existing Claude subscription without separate usage caps of that structure, but depending on the model you are using, you can run out of your credit allocation and you will need to wait a couple of hours.
The user experience differs meaningfully too. ChatGPT Agent is embedded in the familiar ChatGPT chat interface, which reduces adoption friction for existing users. Cowork is a dedicated mode with a cleaner separation between “conversation” and “delegation”, which can actually help teams build better habits around when to use each mode.
Gemini Agent (Google)
Google’s Gemini Agent is the most tightly ecosystem-dependent of the three. It can leverage connected apps like Gmail, Google Calendar, Google Drive, Keep, Tasks, and also Google Maps and YouTube services. If your organization runs on Google Workspace, this integration runs deep, Gemini can manage your inbox, organize Drive folders, schedule calendar events, and execute cross-app workflows without leaving the Google environment.
The tradeoff is flexibility. Gemini Agent is strongest inside the Google ecosystem and progressively less differentiated outside of it. Gemini Agent is currently rolling out on web to Google AI Ultra subscribers in the US, making it the most restricted in availability of the three right now.
For enterprise users already invested in Google Workspace, Gemini’s tight integration with business apps like Salesforce and SAP, available through Gemini Enterprise, is genuinely compelling. For organizations running mixed environments (some Google, some Microsoft, some local tools), that tight coupling becomes a constraint rather than an advantage.
The clearest differentiator for Cowork is its local-first, file-system-native design paired with cross-platform connector flexibility. It doesn’t require you to live in any particular ecosystem. You bring your files, your tools, and your connectors, and Claude works in your environment, not a proprietary one.
Cowork is still labeled a research preview, which means the product will evolve, and some rough edges remain.
The desktop app must stay open while Claude is working. Close it, and the session ends. This matters for longer tasks.
Cowork has no memory across sessions by default. If you want continuity, save context to a file Claude can read at the start of each session, some people use a memory.md file in their working folder for exactly this.
Complex, multi-step tasks consume more of your usage allocation than standard chat. If you’re on a Pro plan, batch related work into single sessions and reserve Cowork for tasks that genuinely need it.
And as with any agentic AI tool: back up important files before granting access. Claude will ask permission before deleting anything, but operating with caution is still the right posture while you’re learning the system.
The most common question I get after people learn about Claude Cowork is: “Okay, but what would I actually use it for?” Fair question. The best way to understand a delegation tool is to see it in action, not in demos, but in the kind of work that eats your Tuesday afternoons.
Here are 30 real examples, organized by the type of work they replace. Each one follows the same pattern: describe the outcome, walk away, come back to finished work.
1. Organize a chaotic downloads folder - You have 200+ files accumulated over months: PDFs, screenshots, installers, random CSVs, renamed drafts. Tell Cowork:
“Organize my Downloads folder. Create subfolders by file type and date. Rename files to be more descriptive. Move anything older than six months to an Archive folder. Don’t delete anything.”
Claude scans every file, reads metadata and content, builds a clean structure, and gives you a summary of what it moved and why. A task that would take hours manually takes minutes.
2. Rename files with consistent naming conventions - Nothing derails version control like “final_FINAL_v3_USE_THIS.pptx.” Point Cowork at a project folder and ask it to rename everything with a consistent format, date prefix, project name, document type, version number. It applies the pattern across every file, handles exceptions intelligently, and produces a log of every change it made.
3. Audit and consolidate duplicate files - Give Cowork access to a large project or archive folder and ask it to identify duplicate files, near-duplicates (same content, different filename), and redundant versions. It produces a report flagging what it found and recommends which to keep, you make the final call before anything is touched.
4. Build a folder structure for a new project - Before a project kicks off, give Cowork your project brief and ask it to create a standardized folder structure. It reads the brief, identifies the major workstreams, creates folders and subfolders, and seeds each with template files (blank meeting notes template, project brief, stakeholder map). Your team starts with a clean setup rather than an improvised one.
5. Build an expense report from receipt screenshots - Drop 40 receipt screenshots into a folder, restaurant meals, hotel charges, Uber rides. Tell Cowork:
“Create an Excel expense report from these receipts. Extract date, vendor, category, and amount. Flag anything where the amount or date isn’t clearly readable as VERIFY. Add a totals row.”
It reads the images, extracts the data, and outputs a fully formatted spreadsheet ready to submit. What used to take 90 minutes takes five.
6. Clean and normalize a messy dataset - Import a CSV from a third-party tool or legacy system and it arrives with inconsistent column names, mixed date formats, blank rows, and rogue special characters. Hand it to Cowork with instructions on what “clean” looks like: standard column headers, ISO date format, no blank rows, specific categorical values. It outputs a cleaned file and documents every transformation it made.
7. Generate a monthly performance report from raw data - If you export raw data weekly from an analytics tool or internal system, Cowork can take those files and produce a formatted monthly summary with totals, trend lines, period-over-period comparisons, all saved as a Word document or Excel file with working formulas. Pair this with a scheduled task and it runs automatically on the first of each month.
8. Convert a data export into an executive dashboard summary - Data teams export detailed reports that executives rarely read in full. Give Cowork the raw export and a brief on what your leadership cares about. It distills the data into a one-page executive summary with the three or four numbers that matter, brief narrative context, and a recommended call-to-action. Formatted, ready to share.
9. Cross-reference two datasets and surface discrepancies - You have a vendor invoice list and an internal purchase order log. They should match. They often don’t. Tell Cowork to compare the two files, match records by invoice number, and produce a discrepancy report showing what’s in one but not the other, and where amounts don’t align. What used to require manual VLOOKUP work becomes a five-minute delegation.
10. Turn meeting transcripts into action items and summaries - Drop six weeks of meeting transcripts into a folder. Tell Cowork:
“Extract all action items, organize by owner, and produce a tracking spreadsheet. Also write a two-paragraph summary of each meeting’s key decisions.”
It reads every file, identifies decisions and commitments, maps them to people, and delivers both the tracker and the summaries. Your meeting archive becomes an operational asset instead of a graveyard.
11. Draft a report from scattered notes - You’ve been collecting notes in a folder for weeks with bullet points, rough paragraphs, half-formed ideas in different files. Tell Cowork to synthesize them into a structured first draft. It reads across all the files, identifies overlapping themes, resolves contradictions, organizes sections logically, and writes a coherent narrative. You edit and refine; you don’t start from scratch.
12. Create a presentation from research materials - You have a folder with research notes, data exports, and a brief. Tell Cowork to produce a ten-slide PowerPoint with a defined structure: executive summary, problem statement, key findings, recommendations, next steps. It plans the structure, writes the content for each slide, and delivers an editable file. Not a static image, a real deck you can open and refine.
13. Produce a weekly team newsletter from Slack exports or meeting notes - Export the week’s Slack highlights or meeting notes into a folder. Tell Cowork to write a concise team update, what shipped, what’s blocked, what’s coming, formatted consistently each week. Add this as a scheduled task and it runs every Friday afternoon before you even remember to do it.
14. Draft a policy document from reference materials - Give Cowork a folder containing reference policies from other organizations, internal guidelines, and notes from stakeholder conversations. Ask it to draft a policy document in a specific format. It synthesizes the inputs, follows your structural template, and produces a formatted draft that reflects the actual content of your source materials, not generic filler.
15. Write job descriptions from role briefs - Give Cowork a folder with a role brief, the team’s existing job descriptions for reference, and any notes from hiring managers. Tell it to produce a draft job description that matches your company’s format and tone. It synthesizes the brief into a structured description and flags anything that seems unclear or inconsistent across your reference files.
16. Synthesize a body of research into a structured report - You have 15 PDFs, academic papers, industry reports, analyst notes, on a topic you need to brief leadership on. Give Cowork access to the folder and a brief on what questions the report should answer. It reads every document, extracts relevant content, identifies patterns and contradictions across sources, and produces a structured synthesis with citations. Deep research that would take a full day takes an hour.
17. Competitive landscape analysis from saved web pages - Save competitor web pages, press releases, pricing pages, and product announcements into a folder using Claude in Chrome. Then ask Cowork to analyze the folder and produce a competitive matrix: key differentiators, pricing positioning, messaging themes, product gaps. It produces a structured analysis document and a comparison tableexactly what you’d want before a strategy session.
18. Audit your own content for consistency - Give Cowork access to a folder of your published content, articles, presentations, proposal templates. Ask it to audit for consistency: terminology, tone, brand claims, messaging pillars. It produces a gap analysis flagging where your content is inconsistent, outdated, or off-brand. Use it before a content refresh or before onboarding a new team member who’ll be producing content.
19. Summarize a book or long document for a specific audience - Upload a long report, academic paper, or book PDF to a folder. Tell Cowork:
“Summarize this for a C-suite audience. One-page maximum. Lead with the three most important implications for enterprise leaders. Use plain language.”
It reads the full document and produces an audience-specific summary, not a generic abstract.
20. Analyze customer feedback or interview transcripts for themes - Drop a folder of customer interview transcripts, survey responses, or support tickets. Tell Cowork to identify recurring themes, surface the most common pain points, count frequency of specific topics, and produce a structured insights report. The output is the kind of synthesis that usually takes a research analyst a full week.
21. Process invoices and extract billing data - Drop a month’s worth of vendor invoices into a folder. Tell Cowork to extract key fields, vendor name, invoice date, due date, amount, line items, and compile them into a master billing spreadsheet. It reads PDFs and image files, handles varying invoice formats, and flags anything where a field isn’t clearly parseable. Accounts payable work without the manual entry.
22. Set up a scheduled weekly briefing - Use Cowork’s /schedule command to create a recurring task that runs every Monday morning: pull from your connected Slack, Gmail, or calendar, synthesize what’s coming up this week, and produce a formatted briefing document in your project folder. You open your laptop Monday morning to a structured overview of the week, already written.
23. Batch-process a set of templates - You have 50 partner organizations that each need a customized version of the same proposal template. Give Cowork the master template and a spreadsheet with partner-specific variables (name, industry, use case, key contacts). It produces 50 customized documents, one per partner, saved to your output folder. What was a half-day mail merge becomes a ten-minute task.
24. Monitor a folder and generate alerts - Set a scheduled Cowork task to check a specific folder daily and send you a summary if new files have appeared that need attention, contract drafts, incoming data exports, flagged documents. It reads the new files, applies your criteria for what counts as “needs review,” and writes you a digest. Lightweight monitoring without a dedicated system.
25. Migrate and reformat a library of documents - You’re switching document templates, new brand guidelines, new format standards, new section structure. Give Cowork access to the old document library and the new template. Tell it to reformat each document to match the new structure. It works through the library systematically, applying the format to each file and producing a conversion log.
26. Prepare a briefing document before a stakeholder meeting - Before a high-stakes meeting, drop relevant background documents, previous meeting notes, project status reports, the other party’s public materials, into a folder. Tell Cowork to produce a one-page briefing: context, what’s been discussed before, open questions, and recommended talking points. You walk in prepared without spending two hours re-reading documents.
27. Draft follow-up communications from meeting notes - After a meeting, drop your notes into a folder and tell Cowork to draft the follow-up email, a summary for the broader team, and individual follow-up messages for each action item owner. It reads the notes, maps commitments to people, writes appropriately calibrated messages for each audience, and saves the drafts. You review and send; you don’t write from scratch.
28. Generate a decision log from a project archive - At the end of a project, give Cowork access to all the meeting notes, email summaries, and documentation in your project folder. Ask it to produce a decision log: what decisions were made, when, by whom, and what information was available at the time. Invaluable for project retrospectives, audits, and onboarding successors.
29. Build an onboarding guide from institutional knowledge documents - Give Cowork a folder of your team’s process documents, SOPs, past project summaries, and style guides. Ask it to synthesize an onboarding guide for a new team member in a specific role. It reads across everything, identifies what’s most relevant to the role, and produces a structured guide that reflects how your team actually works, not a generic HR template.
30. Audit your personal knowledge base for gaps - Drop your notes, research files, and project documentation into a folder. Tell Cowork:
“I’m going to give a keynote on AI adoption in enterprise settings. Review my knowledge base and produce a gap analysis: what topics I have strong material on, what’s thin, and what I should research further before I write the presentation.”
It reads across your files and gives you a map of where your thinking is strong and where it needs development.
You’ll notice these examples share a common structure: they involve multiple files, multiple steps, and output that lands in your file system ready to use. That’s not a coincidence, it’s the signal for when Cowork is the right tool. If a task is a quick question, use Chat. If it involves building something across your files over several steps, delegate it to Cowork.
The fastest way to build the habit is to start with example 1. Take your actual Downloads folder. Give Cowork five minutes to show you what it can do. The rest of the list will start looking very different once you’ve seen it work.
Which of these use cases is most relevant to your current role? I'd love to hear what you try first
.