Claude Cowork expands to web and mobile: most uses are not coding

On Tuesday, Anthropic announced the expansion of Claude Cowork to web and mobile devices, a move that turns its autonomous agent into a platform accessible from anywhere. The beta, initially available to Max subscribers, allows users to start tasks on a laptop, continue them in the background, and review them from a phone, even after closing the app. According to the company, "your work goes everywhere with you and keeps running without you."

The surprise: the majority of usage is not coding

Alongside the launch, Anthropic published data from 1.2 million anonymous Claude Cowork sessions, sampled between May 11 and May 31, 2025, from over 600,000 organizations. The results challenge the dominant narrative that generative AI is mainly used for programming. Only 8.7% of sessions corresponded to software development. Instead, 33.4% went to business processes and operations (such as consolidating reports or creating checklists), and 16.4% to content creation (drafts, presentations, proposals). Together, these two categories account for nearly half of total usage. Other categories include DevOps and infrastructure (7%), research and intelligence (6.4%), data analysis and business intelligence (5.8%), document processing and extraction (4.1%), sales and revenue operations (4%), personal assistance (3.8%), education (2.4%), and meeting intelligence (1.8%).

Anthropic calls these tasks "the work around the work": activities that connect functions and keep projects moving, but rarely appear in job descriptions. "People use it to draft status updates, put together slides, or condense research into a single report," the company explains. This finding is consistent with previous studies by McKinsey and others, which estimate that up to 60% of knowledge workers' time is spent on administrative and coordination tasks, not their core specialty.

Historically, the enterprise AI market has focused on developers, with tools like GitHub Copilot, Cursor, or Replit. However, Anthropic's data suggests that the true mass market lies with knowledge workers: analysts, managers, operations specialists, and content creators. As VentureBeat analyst noted, "this is the biggest secret in enterprise AI." If the trend holds, we will see a wave of products aimed at automating administrative and coordination tasks, not just code.

What changes with the web and mobile expansion?

The update introduces three key capabilities:

• Cross-device synchronization: sessions transfer seamlessly from desktop to mobile and vice versa.
• Background execution: tasks can continue even when the user closes the app, with notifications upon completion.
• Web interface: added to the existing desktop app, making access easier without installation.

This positions Cowork as a persistent assistant, capable of performing long tasks (like analyzing documents or generating reports) while the user handles other responsibilities. Background execution is particularly relevant: it allows complex tasks, such as data reconciliation or multi-source report generation, to run without constant supervision, freeing the user for other activities. This contrasts with traditional conversational assistants that require continuous interaction.

Cross-platform synchronization also addresses a common pain point: work fragmentation across devices. According to a 2024 Gartner report, workers switch devices an average of 10 times a day, and each transition involves a loss of context. Cowork mitigates this by maintaining task state across all devices.

Compared to other AI agents, such as AutoGPT or Microsoft Copilot assistants, Cowork stands out for its focus on "non-technical" tasks and its ability for prolonged autonomous execution. While AutoGPT focuses on programming and technical automation tasks, and Microsoft Copilot is integrated into the Office ecosystem, Cowork seems to target a more generalist niche of "knowledge work" that does not require advanced technical skills.

Implications for the enterprise market

The usage data has profound consequences. Until now, most enterprise AI tools have focused on developers (GitHub Copilot, Cursor, etc.). Anthropic demonstrates that the largest market is among knowledge workers: analysts, managers, operations specialists, and content creators. "This is the biggest secret in enterprise AI," notes VentureBeat analyst. If the trend consolidates, we will see a wave of products aimed at automating administrative and coordination tasks, not just code.

Furthermore, background execution and cross-platform synchronization could turn Cowork into a de facto standard for "AI-assisted asynchronous work." Similar to what happened with Slack for communication or Notion for documentation. However, Anthropic faces competition from giants like Microsoft, which integrates Copilot across its entire ecosystem, and from startups like Adept AI, which develops autonomous agents for enterprise tasks. Anthropic's advantage could be its focus on safety and alignment, values that resonate in corporate environments concerned about privacy.

From a market perspective, Cowork's data suggests that generative AI is following a trajectory similar to personal computing: initially adopted by technical enthusiasts, but eventually finding its greatest utility in general productivity applications. If this trend continues, we could see a shift in enterprise AI investment, from development tools to business process automation platforms.

What should readers know?

For businesses and professionals, the lesson is clear: generative AI is finding its greatest utility in tasks that previously consumed time without being the core of the role. Tools like Claude Cowork allow outsourcing the "organizational glue" (reports, summaries, follow-ups) and freeing up capacity for strategic work. However, it also poses risks: dependence on a single provider, data privacy (by delegating sensitive tasks to the cloud), and potential loss of synthesis and writing skills. Companies must carefully evaluate which tasks to delegate and establish usage policies that protect confidential information.

Additionally, the expansion to web and mobile could accelerate adoption in sectors like consulting, finance, and human resources, where professionals need access to tools from anywhere. However, reliance on internet connectivity and latency in complex tasks remain technical challenges to solve.

In terms of competition, Anthropic faces OpenAI, which with ChatGPT Enterprise also targets knowledge workers, but with a more conversational approach. Anthropic's bet on autonomous and persistent agents could differentiate it, as long as it maintains accuracy and avoids hallucinations in critical tasks. The published data is encouraging, but it is based on anonymous sessions and does not reveal error rates or user satisfaction.

"AI does not replace experts, but it can replace the work that surrounds expertise," Anthropic concludes. And the numbers back it up. For business leaders, the recommendation is to closely monitor this evolution and consider controlled pilots to assess the real impact on productivity and work quality.