Anthropic Launches Claude Cowork for Web and Mobile: Your AI That Keeps Working

Anthropic Launches Claude Cowork for Web and Mobile: Your AI That Keeps Working

Anthropic announced on July 7, 2026 that Claude Cowork is coming to web and mobile — bringing its autonomous AI agent capabilities to platforms beyond the desktop application. The announcement positions Cowork as an AI that "keeps going without you," continuing complex tasks independently after a user steps away.

This is a meaningful expansion of Claude's capabilities from conversational AI to autonomous agent — one of the most actively contested areas in AI development as companies race to build AI that can handle multi-step tasks without constant human input.

---

What Is Claude Cowork?

Claude Cowork is Anthropic's agentic AI feature that allows Claude to continue working on tasks after the user has closed the app or stepped away from their device. Rather than pausing when the user leaves, Cowork maintains a persistent work session — executing tasks, making decisions within defined boundaries, and delivering completed work when the user returns.

The concept addresses one of the core limitations of current conversational AI: the need for the human to remain in the loop for every step. Claude Cowork's tagline captures the vision directly: "Your work goes everywhere with you, and keeps going without you."

---

Why This Matters: From Chat to Agent

The distinction between a conversational AI (like standard ChatGPT or Claude) and an agentic AI (like Claude Cowork) is fundamental:

For professionals with complex, multi-step work — research synthesis, code review, document analysis, data processing — Cowork represents a step change from AI as a tool you use to AI as a colleague you delegate to.

---

Cowork on Web and Mobile: What's New

The July 7 announcement specifically addresses the expansion of Cowork to web and mobile platforms:

Mobile (iOS and Android):

• Users can initiate a Cowork task from their phone and check back later for results
• Push notifications when a Cowork session completes or needs input
• Lightweight mobile interface for reviewing completed work
• Ability to delegate tasks during commutes or breaks

Web (Browser):

• Full Cowork workflow accessible from any browser without the desktop app
• Shared Cowork sessions for teams (a colleague can hand off a Cowork task to you)
• Integration with web-based tools in Cowork's connector ecosystem

This accessibility expansion is significant: Cowork previously required the Claude desktop application, limiting its use to desktop/laptop contexts. Web and mobile access means Cowork can be initiated and monitored from any device at any time.

---

Claude Reflect: Beta Feature for Personalisation

Alongside the Cowork announcement, Anthropic also introduced "Reflect" on July 9, 2026 — a beta feature that gives users a new way to understand and refine how they use Claude.

Based on Anthropic's research interviews, a recurring user request was for Claude to adapt its behaviour based on accumulated experience of how a specific user works — their preferences, communication style, and recurring task patterns.

Reflect addresses this by:

• Surfacing patterns in how you use Claude over time
• Suggesting refinements to how Claude responds to you
• Allowing you to set persistent preferences based on what you've learned works well

This personalisation layer is a direct response to the "memory problem" in AI — making AI assistants that feel like they know you rather than starting from scratch each session.

---

The Safety Context: Anthropic's Approach to AI Agents

Anthropic's approach to autonomous AI agents reflects its safety-first philosophy. Cowork is designed with what Anthropic calls "minimal footprint" principles:

• Explicit permission scope — Cowork operates only within the permissions and tools a user explicitly grants
• Checkpoint prompts — For consequential actions (sending emails, making purchases, modifying files), Cowork requests explicit confirmation before proceeding
• Reversibility preference — Where possible, Cowork prefers actions that can be undone
• Transparency log — Every action Cowork takes is logged and visible to the user

These principles differentiate Cowork from more autonomous AI agents that take broader, less reversible actions — reflecting Anthropic's mission focus on building AI that is both capable and safe.

---

Claude Enterprise: New Analytics and Cost Controls

On the same date, Anthropic also announced new analytics and cost control features for Claude Enterprise customers:

• Usage dashboards showing consumption by team, project, and user
• Spend caps and budget alerts for enterprise deployments
• API cost attribution linking usage to specific business workflows
• Model preference controls for organisations that want to lock specific teams to specific Claude models

These features address a common enterprise AI complaint: the "shadow AI" problem where individual employees rack up significant AI costs without visibility or governance from IT and finance teams.

---

How Claude Cowork Compares to Competitors

The autonomous AI agent space has become intensely competitive in 2026:

Anthropic's positioning centres on trustworthiness — building an agent that enterprise users and individuals can delegate work to with confidence that it won't take unintended or irreversible actions.

---

Practical Use Cases for Claude Cowork

Based on Anthropic's guidance and early user reports, Claude Cowork is most effective for:

Research and analysis:

• "Research the top 10 competitors in the SMB CRM market, find their pricing pages, and compile a comparison table"
• Cowork browses the web, extracts information, and delivers a completed document

Document processing:

• "Review these 50 contracts, flag any non-standard indemnification clauses, and summarise what you find"
• Cowork reads each document and delivers a structured summary

Code review:

• "Review this codebase for security vulnerabilities, document findings, and suggest fixes for the top 5 issues"
• Cowork analyses code files and delivers a security report

Content pipeline:

• "Turn these 10 research papers into executive summaries in plain English"
• Cowork processes each paper and delivers formatted summaries

---

What's Not Yet in Cowork

It's worth noting what Cowork doesn't yet support in this launch:

• Direct action execution — Cowork can research and draft, but can't yet book meetings, make purchases, or send emails without explicit approval at each step (by design, per safety principles)
• Real-time collaboration — Multiple users working with the same Cowork session simultaneously is not yet available
• Integration with all third-party apps — The connector ecosystem is growing but doesn't yet cover all major business tools

---

Availability and Pricing

Claude Cowork is available to:

• Claude Pro subscribers ($20/month) — individual use
• Claude Max subscribers ($100/month) — extended Cowork usage limits
• Claude Team and Enterprise — team collaboration features, admin controls, usage dashboards

Web and mobile availability is rolling out through July 2026. Users can access Cowork at claude.ai from any browser, and through the updated Claude apps on iOS and Android.

---

Frequently Asked Questions

Is Claude Cowork safe to give access to my files and data? Anthropic has designed Cowork with explicit, minimal-footprint principles. It only accesses data and tools you explicitly grant it permission to use, and logs every action it takes. That said, any AI with access to sensitive data carries risk — review what permissions you grant carefully.

How does Cowork handle tasks it's unsure about? By design, Cowork stops and asks for clarification rather than guessing on ambiguous or consequential decisions. This is a core safety principle — preferring a checkpoint pause over potentially wrong autonomous action.

Can my team see my Cowork sessions? On Team and Enterprise plans, administrators can see usage analytics. Specific session content visibility depends on your organisation's Claude Enterprise settings. Individual Pro and Max subscribers have private sessions.

What happens if Cowork makes a mistake? Cowork's design preference for reversible actions means most mistakes can be undone. The action log lets you see exactly what it did. For consequential actions, Cowork asks for confirmation before proceeding — this is the primary safeguard against significant errors.

How is this different from Claude Projects? Claude Projects maintains persistent memory and context across conversations. Cowork takes autonomous action — it doesn't just remember your preferences; it actively continues working on tasks after you've stepped away. The two features complement each other.

---

Conclusion

Claude Cowork's expansion to web and mobile is part of a broader industry shift: from AI as a responsive tool to AI as an active participant in knowledge work. Anthropic's safety-first approach differentiates Cowork in a market where competitors are pushing for maximum automation — betting that enterprise users will ultimately trust an AI that's careful more than one that's fast.

For professionals who regularly spend hours on research, document analysis, and content synthesis, Cowork represents a meaningful productivity step. The combination of Cowork's autonomous task completion with Claude's industry-leading long-context understanding positions it as one of the more practical agentic AI products available today.

Access Cowork at claude.ai — the web version is available to all Pro and above subscribers starting today.