Anthropic has expanded Claude Cowork beyond the desktop. The agentic workspace is now available on web and mobile platforms. The updated version fundamentally changes how users interact with the AI, allowing tasks to persist across devices and continue running even when the user is offline.
The beta rollout has commenced for Max-tier subscribers, with access expanding to other plans over the coming weeks. To mark the launch, Anthropic has extended its doubled Cowork usage limits through the fifth of August.
The reality of AI usage in the workplace
AI agents initially gained traction through software development. However, recent data from Anthropic paints a very different picture of actual enterprise usage. Following an analysis of 1.2 million anonymised sessions across more than 600,000 organisations, the company revealed that over 90 per cent of Cowork activity involves non-coding tasks.
The largest categories of use are business operations at 33.4 per cent and content creation at 16.4 per cent. These tasks include reconciling spend, drafting variance memos, building client presentations and tracking contract renewals. Together, these two categories account for roughly half of all Cowork usage.
Anthropic framed this insight perfectly in their launch announcement. "It's the work around the work: rarely in anyone's job description, but a large share of everyone's week."
That framing resonates strongly with project delivery professionals. The administrative overhead of project management often consumes more time than the core delivery work itself. Reconciling invoices, consolidating status reports, updating risk registers and managing stakeholder communications occupy significant portions of the project manager's week.
Untethering the workflow
The expansion to web and mobile introduces three significant architectural shifts to the Cowork experience.
Cross-device continuity: Sessions and files sync across desktop, web and mobile. A user can initiate a task on a laptop and monitor its progress from a smartphone. The session maintains its full context and state across devices. This is particularly valuable for project managers who need to check task progress whilst on-site or in meetings.
Offline background execution: Scheduled tasks no longer require an active device connection. A task queued for the early hours of the morning can process emails, compile briefings and draft responses entirely unattended. A project manager can schedule a task to run at 6 a.m., and by the time they arrive at the office, the preliminary analysis is complete and ready for review.
Human-in-the-loop checkpoints: When Claude encounters a decision requiring human judgement, it pauses and sends a query to the user's phone. No final actions are taken without explicit approval, which allows human control over critical decisions whilst allowing the agent to handle routine work autonomously.
Armmand Hosseini, Customer Success at Ramp, shared a practical example of this flexibility. "I built a dashboard to track my clients while travelling. I started on my laptop and picked the session up on my phone while waiting for my bag to come out. It just held the thread."
Desktop remains the primary hub
Whilst the web and mobile versions offer unprecedented flexibility, the desktop application remains the most capable environment. Only the desktop client can access local files and interact directly with the user's browser. The new platforms serve to extend the reach of Cowork to users who cannot install desktop applications or who require mobile monitoring.
On both web and desktop, chat and Cowork sessions now share a unified interface. Projects and artefacts are synchronised across both surfaces, reducing the need for context-switching. A project manager can start a conversation in chat and transition to a Cowork task without losing context.
Automating project management administration
For project delivery professionals, the "work around the work" often consumes the majority of the week. The ability to schedule background tasks to compile reports, review meeting transcripts and update risk registers overnight is highly valuable.
Consider a typical scenario. A project manager has spent the day in meetings and site visits. They have collected dozens of photographs, voice notes and email threads. Rather than spending the evening manually consolidating this information, they can delegate the task to Claude Cowork. The agent can process the photographs, transcribe the voice notes, extract key information from emails and compile a comprehensive project status update.
By morning, the project manager has a draft status report ready for review. They can approve it, make minor edits and send it to stakeholders. What would have taken two hours of manual work is now completed overnight.
Security and approval gates
We recognise that maintaining a human-in-the-loop for final approvals addresses many of the security concerns associated with autonomous agents. This approach allows project managers to leverage automation whilst retaining control over client communications and critical project data.
The approval gate model is particularly important for AEC firms. Client communications must be carefully reviewed before sending. Contract changes must be approved by appropriate stakeholders. Risk escalations must be reviewed by senior project leadership. By requiring human approval before any final action, Cowork maintains the governance structures that AEC firms require.
The broader implications for knowledge work
The data from Anthropic suggests that the AI industry has been focused on the wrong use case. The narrative around AI agents has centred on software development and coding tasks. However, the actual usage data shows that the greatest value lies in automating administrative and operational work.
The implications are profound for how AEC firms should think about AI adoption. Rather than looking for ways to automate technical delivery work, firms should focus on automating the administrative overhead that surrounds delivery. The return on investment is likely to be higher, and the implementation is likely to be simpler.
Takeaway
• The vast majority of AI agent usage is currently focused on business operations and content creation, not software development, suggesting that the greatest value lies in automating administrative work.
• Background execution allows complex administrative tasks to be processed overnight without requiring an active device connection, enabling project managers to focus on higher-value work during business hours.
• Cross-device synchronisation enables professionals to monitor and approve agentic workflows whilst on the move or on-site, maintaining flexibility and responsiveness.
• Human-in-the-loop checkpoints ensure that critical decisions and communications always require explicit user approval before execution, maintaining governance and control.
• The unified interface between chat and Cowork sessions reduces context-switching and allows seamless transitions between conversational assistance and agentic task execution.
Subscribe to the Project Flux newsletter to explore how AI agents are transforming daily project operations and administrative workflows.
Links and Stuff
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
1