Claude Cowork is now accessible on web and mobile, in addition to desktop, with a beta rollout starting with the Max plan. For an SMB, the change is tangible: an AI agent can continue background tasks, retrieve its files from one session to the next, and work across multiple devices. But the full experience remains tied to Claude Desktop for local access to the comporter.
Claude Cowork on web and mobile: what really changes
Anthropic launched Claude Cowork on January 12, 2026, as a research preview, initially on Claude Desktop for macOS and for Max subscribers. On January 16, access was extended to Pro users on macOS. The new development in July 2026 is the arrival of Claude Cowork on web and mobile, with a beta rolled out gradually, starting with Max.
The key takeaway: Cowork is not just a chat window. Anthropic presents it as an intellectual work agent, meaning an assistant capable of chaining together multiple actions to produce a result, similar to Claude Code but not limited to software development. It is therefore aimed at office use cases: summaries, document preparation, file analysis, research, task coordination.
For a business leader, the benefit is not to “do AI” just to check a box. It is to shift part of the repetitive or preparatory work to a tool that can operate in the background. A simple example: request an analysis of customer feedback in the morning, resume the session on mobile between two meetings, then finalize it on a comporter with the files already available.
Remote sessions: the practical advantage, and the hidden limitation
Claude Cowork works by default with remote sessions in beta. In concrete terms, the agent loop, meaning the mechanism that decides the next actions, and code execution run on the servers Anthropic’s. Sessions and files are saved in the user’s Claude account, making it possible to retrieve them on desktop, web, and mobile.
This is useful for long tasks. Anthropic indicates that work can continue in the background, that scheduled tasks can run even without a device online, and that the same elements remain available across platforrms. For a management team, this brings AI closer to an asynchronous collaborator rather than a simple tool consulted occasionally.
The limitation is less visible. The full experience remains on desktop. Access to local files, browser use, and control of the comporter from web or mobile require the Claude Desktop application to be open on your machine. In other words, mobile is very effective for monitoring, giving instructions, and resuming a session; it does not yet fully replace the workstation.
In the projects we lead, we often see the same misunderstanding with AI tools: the gain comes less from the general promise than from the way the workflow is structured. If your documents are scattered across Google Drive, Microsoft 365, a CRM, and local files, the agent will be useful only if access, rules, and responsibilities are defined before deployment.
Plugins, connectors, and SMBs: where the value lies
On January 30, 2026, Anthropic added plugin support to Claude Cowork. According to TechCrunch, Anthropic also opened 11 internal plugins on that occasion, and Anthropic’s Matt Piccolella explained that these plugins were designed to be customizable for businesses. The principle is simple: connect the agent to business tools instead of copying and pasting everything by hand.
In May 2026, Anthropic announced Claude for Small Business, with connectors and workflows that can be activated in Claude Cowork. The cited integrations include Intuit QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365. For an SMB, these names speak more clearly than technical debates: invoicing, payments, CRM, documents, signatures, office productivity.
The right question therefore becomes: which tasks are worth entrusting to an agent? Honestly, automating a rare and sensitive operation is not always cost-effective. By contrast, preparing a weekly sales tracking dashboard from HubSpot and Google Workspace, or producing a first draft of a client file from approved documents, can save hours without changing your information system.
This shift aligns with a broader trend: AI agents do not just replace software, they change the way work loops are delegated. To understand this logic, the analysis on loop engineering applied to AI provides a useful framework before connecting tools everywhere.
Security and confidentiality: what to check before using Cowork
Anthropic specifies that remote sessions use isolated temporary environments, called sandboxes (separate spaces), on infrastructure managed by Anthropic. The documentation also indicates that there is no default access to private or internal networks. That is reassuring, but it is not an exemption from governance.
The trap that non-technical people underestimate: the risk does not come only from a server vulnerability. It also comes from a poorly defined access scope. An agent connected to Google Workspace or Microsoft 365 can handle very broad documents if user permissions are too generous. Before testing Cowork on client data, check permissions, activity logs, and retention rules.
Visit RGPD, applicable since 2018, imposes a clear framework: purpose, data minimization, information of individuals, and control of subcontractors. If you use Claude Cowork to process personal data, you need to know what data enters the tool, why, who has access to it, and how long it is kept. It is administrative, yes. But it is often less costly than a correction after an incident.
Teams that also develop mobile applications must maintain the same discipline on the product side. The principles of privacy by design for a mobile application apply very well to AI agents: reduce exposed data, segment access, and anticipate human errors.
Costs, timelines, and trade-offs for a business project
The exact pricing of Claude Cowork depends on the Claude plans available at the time of deployment, and the web/mobile beta starts with Max according to Anthropic. The real cost of adoption for an SMB is not limited to the subscription, however. You also need to account for scoping, access rights, testing, troring, and sometimes the creation of workflows or plugins.
With that budget, it is better to start with a limited pilot rather than a general deployment. A two- to four-week pilot is often enough to know whether the agent really reduces the time spent on a task. The metric should be simple: time saved, errors avoided, production lead time, and satisfaction of the users concerned.
| SMB scenario in 2026 | Realistic timeline | Indicative budget for the French market | Point of vigilance |
|---|---|---|---|
| Individual test of Claude Cowork on office tasks | 1 to 3 days | Relevant Claude subscription, hors support | Do not use sensitive data at the start |
| Team pilot with Google Workspace or Microsoft 365 | 2 to 4 weeks | Around €2,000 to €8,000 depending on scoping and formation | Access rights, internal procedures, measurement of gains |
| Business workflows with CRM, signature, or payment connectors | 4 to 8 weeks | Around €8,000 to €25,000 depending on complexity | Legal validation, human takeover, logs |
| Plugin or custom integration | 6 to 12 weeks | Often €15,000 and more depending on API and security | Maintenance, API limits, non-regression testing |
These orders of magnitude are not Anthropic rates. They correspond more to what a French SME may encounter in scoping, integration, security, and change management services. The most underestimated item remains maintenance: a useful workflow today can break if an API changes, if a connector evolves, or if internal rules become stricter.
For projects where AI must be integrated into a client application or a business tool, Cowork is not necessarily the right starting point. A AI mobile application for SMEs may be more suitable if the use case needs to be distributed to clients, field technicians, or franchisees. Cowork is above all an internal productivity tool.
When Claude Cowork is not the right answer
The obvious solution can be the wrong one if the process is not stabilized. An AI agent quickly amplifies ambiguities: unclear instructions, contradictory data, poorly defined responsibilities. Before automating, it is sometimes better to simplify the process itself.
Another delicate case: decisions with legal, financial, or HR impact. Claude Cowork can prepare, summarize, and compare, but validation must remain human. DocuSign, PayPal, QuickBooks, or HubSpot handle binding actions; an error can cost more than the time saved.
Here are the checks to plan for before serious initial professional use:
- Choose a precise, frequent, and measurable use case rather than a “general assistant.”
- Create a test account or space without critical documents at the start.
- Limit permissions to the necessary folders and tools.
- Plan for human validation before any external action: sending, signing, payment, CRM modification.
- Document reusable instructions to avoid each user improvising.
From an agency perspective, the instinct is to treat this type of agent like a mini product project: internal persona, journey, risks, success criteria, and only then technical integration. This method avoids confusing an impressive demonstration with lasting value.
If your challenge also involves web visibility, agentic AI raises another question: are your content and data structured enough to be understood, cited, or properly reused? Reflections on content cited by AI and the internal linking understandable by Google and the models complement an internal automation approach well.
Defining this type of project upstream avoids most unpleasant surprises: overly broad access, unrealistic expectations, gains that are impossible to measure. An external perspective is especially helpful in choosing the right first use case, the one that proves the value without unnecessarily exposing the company.
FAQ about Claude Cowork
Is Claude Cowork available on iPhone and Android?
Anthropic states that Claude Cowork is available on web, desktop, and mobile, with a web/mobile beta rollout starting with the Max plan. Exact availability may vary depending on accounts and plans.
What is the difference between Claude Cowork and Claude Code?
Claude Code is primarily aimed at software development. Claude Cowork follows an agentic logic as well, but for general knowledge work: documents, research, analyses, workflows, and office tasks.
Can Claude Cowork access my local files?
Yes, but the full experience remains tied to Claude Desktop. For access to local files, the browser, or the cordinateur from the web or mobile, the desktop application must be open on your machine.
Is Claude Cowork suitable for an SME?
Yes, if the use case is specific and access is properly controlled. For an SME, the first gains often come from recurring tasks: summaries, document preparation, sales follow-up, or information consolidation.