Claude Cowork Hits Mobile: Why Content Creation Won the Agentic Race

When Anthropic expanded Claude Cowork to mobile and web for Max subscribers this week, it released the first real usage breakdown for its agentic task runner — and the data contradicts the dominant narrative about who benefits most from agentic AI. Business process automation leads at 33.4%. Content creation comes in second at 16.4%. Software development — the category every vendor has been optimizing for — sits at 8.7%.

For content teams, this is validation. For the agentic AI market, it's a reframe. The "AI coding assistant" story that dominated 2024 and 2025 turns out to be a fraction of the actual usage. The teams doing unstructured, judgment-heavy work — running campaigns, managing briefs, coordinating across channels — are the ones finding the most daily utility in autonomous AI agents.

What Claude Cowork Actually Does

Claude Cowork is Anthropic's asynchronous agent system — a layer on top of Claude that can receive a task, break it into steps, execute those steps using connected tools, and complete the work without the user staying in the conversation. The mobile and web expansion means users can kick off a task from their phone and check back on progress without keeping a laptop session active.

The distinction from a standard Claude conversation is significant for content teams. A chat-based AI workflow requires you to be present and steering throughout. Cowork accepts the task, executes it, and reports back — closer to delegating to a contractor than operating a tool. For content workloads that have clear inputs and outputs but require multiple steps and judgment calls along the way, that model fits naturally.

Why Content Creation Outperforms Coding in Agentic AI

The common assumption was that software development would dominate agentic AI — it's structured, verifiable, and has clear success criteria. What the Cowork data suggests is that this advantage doesn't survive the switch from synchronous to asynchronous work. When you're coding, you need to evaluate each step before the next. The feedback loop is too tight for delegation to work well without constant presence.

Content creation has a different feedback structure. The inputs are a brief, a tone guide, a target keyword, and an outline. The output is a draft that you review at the end — not a step-by-step process you supervise. This is exactly the shape of work that asynchronous agents handle well. You define the task, come back later, and the work is done.

The same logic explains why business process automation leads: invoice processing, scheduling, status reporting, data extraction from documents — these tasks are well-defined, have clear completion states, and don't require the human to be in the loop at each step. Content production tasks share these properties in ways that open-ended coding tasks don't.

What This Means for Content Team Workflows

If content creation is already the second-largest use case for enterprise agentic AI, content teams that haven't built Cowork-style asynchronous workflows are behind the adoption curve — not ahead of it. The early adopters have already moved from prompt-and-review to delegate-and-collect. Here are the workflow patterns that map best to agentic execution:

Brief-to-draft pipelines

The highest-value agentic content workflow is brief ingestion → research → outline → draft, executed without human checkpoints. You provide a structured brief — target keyword, audience persona, content type, competitor URLs to reference — and the agent handles the rest. The output is a complete first draft ready for editorial review, not a starting point that still requires hours of assembly.

Content repurposing queues

Repurposing is one of the highest-volume, lowest-judgment tasks in content operations — turning a long-form article into LinkedIn posts, email copy, a tweet thread, and a short video script. This is pure execution against a template. Agentic systems can process a queue of repurposing tasks overnight and deliver a complete batch by morning.

Research digests and competitive monitoring

Gathering competitor content updates, monitoring for trending topics, and surfacing relevant news to brief writers is time-consuming and routine. Cowork-style agents are well-suited to this type of scheduled reconnaissance — run nightly, deliver a digest by morning, flag items that require editorial action.

Content ops administration

The business process automation category (33.4%) captures a large slice of what content teams actually spend time on: updating content calendars, tracking publication status, managing editorial feedback loops, updating metadata across a CMS. These tasks are high-volume, low-value individually, and exactly what agentic delegation eliminates.

The Mobile-First Agentic Workflow

The expansion to mobile matters because it completes the delegation model. The previous constraint was that you needed to be at a desktop to initiate Cowork tasks — which defeated much of the point. Now you can queue a batch of content jobs from your phone in the morning, and they'll be done before you open your laptop.

The cross-device continuity also enables a new pattern: asynchronous review. If you delegate ten content tasks before a meeting, you can review the completed drafts on your phone during your commute home. The production and review loop runs in parallel with the rest of your day rather than blocking it. Content teams that internalize this pattern will see their effective output multiply without extending working hours.

Conclusion

The Claude Cowork usage data tells a story that should reshape how content teams think about agentic AI investment. Content creation is not a secondary use case for enterprise AI agents — it's the second-largest one, ahead of the category that commanded most of the industry attention. Content teams are already the primary beneficiaries of this infrastructure. The question is whether your team has built workflows that take advantage of asynchronous, delegated execution — or whether you're still treating every AI interaction as a synchronous, present-in-the-loop task.