For two years, the most capable AI agents lived in the terminal. Tools like Claude Code could plan multi-step work, run commands, and edit files autonomously — but you needed to be a developer to benefit. That era just ended. On July 7, 2026, Anthropic expanded Claude Cowork from its desktop-only debut to web and mobile, packaging the same agentic engine behind Claude Code into an interface built for people who have never opened a terminal in their lives.
The timing tells the real story. Anthropic's own usage data shows that over 90 percent of Cowork sessions are not coding at all. They are finance reconciliations, HR onboarding checklists, status reports, and spreadsheet cleanups. The coding agent wars have officially spilled into the rest of the office — and if you run a business, this is the moment to pay attention.
What Is Claude Cowork?
Claude Cowork is Anthropic's autonomous work platform for knowledge workers. It first appeared as a desktop app in January 2026, and since July 7 it runs on web and mobile for Max subscribers, with sessions executing in the cloud by default.
The core difference from a chatbot is simple: Cowork does not describe how to do a task — it does the task. You give Claude permission to read, edit, and create files in folders you specify, and it directly manipulates them. It structures folders, builds presentations, and writes spreadsheets with working formulas, rather than printing text for you to copy somewhere else.
Under the hood, each task follows a consistent loop:
- Plan — Claude analyzes your request and drafts a plan, breaking complex work into subtasks when needed.
- Execute — it runs code and shell commands in an isolated environment on Anthropic's servers, working through the subtasks.
- Check in — when it hits a decision only you can make, it pauses and asks instead of guessing.
- Deliver — finished files land in your designated folders, ready to use.
Cloud Sessions: Close the Laptop, Work Continues
The July update's headline feature is cross-device continuity. Cowork sessions now run in the cloud by default, which changes the daily experience in three concrete ways:
- Start anywhere, finish anywhere. Kick off a task at your desk, check its progress from your phone at lunch, and collect the finished output in the evening.
- Offline execution. When you shut your laptop, tasks keep running on Anthropic's infrastructure. Scheduled work runs even if none of your devices are online.
- Push-notification approvals. When Claude needs a decision — approving a draft, choosing between two options, greenlighting a send — it pings your phone. Nothing ships without your sign-off.
That last point matters more than it sounds. The biggest failure mode of autonomous agents in business settings is silent wrong turns, a problem we covered in depth in why AI agent projects fail without a human in the loop. Cowork bakes the human checkpoint into the product rather than leaving it to policy.
What People Actually Use It For
Anthropic published a breakdown of Cowork usage alongside the launch, and it reads like a job description for back-office operations. The largest category, at 33.4 percent of sessions, is business process operations: pulling scattered updates into a single report, building onboarding checklists, and reconciling spreadsheets. The heaviest users work in finance, HR, and administration — not engineering.
Typical tasks that fit the pattern:
- Turning a folder of contracts into a renewal tracker with dates and owners
- Reconciling two exports from different systems into one clean spreadsheet
- Drafting a weekly status report from meeting notes and project files
- Converting a messy inbox thread into a structured action list
- Building a first-draft slide deck from a written brief
Microsoft 365 Write Access: The Quiet Big Deal
Cowork has connected to Microsoft 365 through an MCP connector since February 2026, but only with read access to Outlook, OneDrive, SharePoint, and Teams. In July 2026, Anthropic flipped the switch on write capabilities: Claude can now draft and send emails, manage calendar events, and create or update files across the M365 suite.
For businesses running on Microsoft 365 — which describes most SMEs in the MENA region — this closes the loop. An agent that can read your files but not act on them saves you reading time. An agent that can reconcile the spreadsheet, file it in the right SharePoint folder, and email the summary to your team saves you the whole afternoon.
Cowork vs. ChatGPT Work
Two days after the Cowork expansion, OpenAI answered with ChatGPT Work, a new mode powered by GPT-5.6 that folds Codex into a rebuilt desktop app. The two products are converging on the same promise from different directions:
| Dimension | Claude Cowork | ChatGPT Work |
|---|---|---|
| Origin | Claude Code's agentic engine, repackaged | ChatGPT plus Codex, merged |
| Strength | Local and cloud file pipelines — analyze a spreadsheet, build the deck, draft the email | Web-based workflows — research, booking, multi-site data gathering |
| Integrations | Microsoft 365 MCP connector with read and write access | Microsoft 365, Google Drive, Slack, Notion |
| Availability | Max subscribers, desktop, web, and mobile | Pro, Enterprise, and Edu, with Plus and Business following |
| Oversight | Push-notification approvals on your phone | Enterprise governance controls and real-time monitoring |
The honest read: if your work lives in files — Excel, PowerPoint, contracts, reports — Cowork's direct file manipulation is the stronger fit today. If your work lives across websites and third-party apps, ChatGPT Work's browsing-first design has the edge. Many teams will end up running both, a pattern we recommend planning for in our guide to multi-model vendor resilience.
What This Means for MENA Businesses
The strategic shift is that agentic AI no longer requires a development team to deploy. Until now, getting agent-grade automation into a Tunisian or Saudi SME meant custom integration work. Cowork collapses that to a subscription and a folder permission.
Three practical recommendations:
- Start with one repetitive file workflow. Pick a monthly reconciliation, a recurring report, or a document-heavy process. Give Cowork a scoped folder and evaluate the output against your current baseline for a full cycle.
- Keep the approval gate on. The push-notification checkpoint is the product's best safety feature. Resist the temptation to rubber-stamp; the first month of reviews is where you learn what the agent gets wrong.
- Inventory your data boundaries first. Cowork executes in Anthropic's cloud. Before pointing it at client files, verify what your contracts and local regulations permit — the same due diligence we outline in our AI agent readiness playbook for MENA SMEs.
The Bottom Line
Claude Cowork's expansion to web and mobile is not just a platform update — it is the clearest signal yet that autonomous agents are becoming standard office equipment. The 90 percent non-coding usage figure settles the question of who agentic AI is for: it is for everyone who works with files, and that is essentially everyone.
The window where adopting this class of tool counts as an advantage rather than table stakes is open now, and it will not stay open long. If you want help mapping which of your workflows are ready for an agent — and which need guardrails first — talk to our team. Automating business operations with AI is exactly what we do.