Anthropic integrates always-on Cowork agents into Claude

Anthropic has integrated its always-on agent feature, previously known as Cowork, directly into the main Claude application. This update, released in September 2026, allows Pro and Max subscribers to delegate recurring tasks that continue running after they log off. The move positions Claude against recent launches from competitors, including OpenAI’s Dots and Meta’s Muse, as companies vie for user trust in autonomous AI workflows.

What happened

The integration marks a strategic shift for Anthropic, folding a capability that had been available since July into its primary interface. Unlike rivals who launched standalone products with distinct branding, Anthropic updated its existing platform without a new name or mascot. The feature enables users to schedule jobs that execute in the cloud, independent of the user’s local device status. While Team and Free plan access is pending, current paying customers can immediately leverage these persistent agents.

This development occurs amidst a rapid cycle of feature replication among major AI labs. OpenAI introduced Dots at its DevDay conference, offering always-on agents with dedicated cloud computers and browser access. Meta’s Muse, which launched earlier, has seen significant mobile adoption, while xAI rolled out Grok Bot in August. Anthropic’s approach differs by targeting developers and teams already embedded in the Claude ecosystem, prioritizing deep integration over broad consumer appeal. The company aims to convert existing users into heavy adopters of automated workflows rather than chasing viral download metrics.

Competitive dynamics show that features spread quickly across platforms. For instance, OpenAI’s Work mode, launched in July 2026, was mirrored by Anthropic’s merge two months later. However, Anthropic’s underlying Cowork technology actually shipped on July 7, just before OpenAI’s public release. Internal data suggests strong adoption among engineering teams, with one internal version of Claude Tag accounting for approximately 65% of code changes within product teams. This indicates that while consumer apps grab headlines, developer utility drives sustained usage.

How it works

The core mechanism involves cloud-based execution environments that persist beyond a user’s session. When a user assigns a scheduled task, the agent operates on remote infrastructure, maintaining state and access permissions independently. This contrasts with local bots that require a personal computer to remain powered on. The system relies on specific identity management, such as Claude Tag in Slack, which operates with its own service identity and scoped access controls. This ensures that agents act within defined boundaries rather than sharing a single set of logins across an entire organization.

Trust and security are central to the design. Anthropic defaults to a cautious stance, requiring explicit confirmation before taking actions unless configured otherwise. Each agent maintains a detailed audit trail, allowing users to review decisions and interventions. This transparency addresses common concerns about autonomous agents holding credentials and acting without supervision. By integrating these safeguards directly into the workflow, Anthropic aims to reduce the friction associated with delegating critical tasks to AI.

Key details

• Anthropic merged the Cowork feature into the main Claude app in September 2026.
• The update is currently available to Claude Pro and Max subscribers, with Team and Free plans coming soon.
• Claude Tag, available in Slack since June 2026, provides proactive assistance with its own identity and audit trail.
• Internal usage data shows Claude Tag accounts for about 65% of code changes in some product teams.
• Meta’s Muse achieved an estimated 359,000 daily U.S. iOS users in its first 12 days, surpassing ChatGPT’s early numbers.
• OpenAI’s Dots require a ChatGPT Pro plan, costing between $100 and $500 per month, due to high compute demands.

Why it matters

For software engineers and technical leads, the shift toward always-on agents changes how maintenance and monitoring tasks are handled. Instead of relying on manual checks or rigid scripts, teams can delegate complex, multi-step workflows to agents that understand context. This reduces the operational burden of nightly QA, invoice checking, or bug triage. However, it also introduces new dependencies on AI reliability and security posture. Understanding the distinction between shared login models and scoped identity agents is crucial for maintaining secure infrastructure.

The competition highlights a divergence in strategy between consumer-focused and developer-focused AI products. Meta relies on its massive user base to drive adoption, while Anthropic and OpenAI focus on high-value, paid users who integrate AI into professional workflows. For builders, this means the tools are becoming more robust but also more expensive. The real test is not just whether the agent can start a job, but whether it can complete it without excessive human intervention. Trust is the currency of this new era, and auditability is the proof.

What you can do

• Evaluate your current recurring tasks, such as log analysis or report generation, for potential automation via Claude Cowork.
• Review the permission scopes of any AI agents you deploy, ensuring they use scoped identities rather than shared admin credentials.
• Monitor the audit trails provided by tools like Claude Tag to verify agent actions and refine prompt instructions.
• Compare the cost-benefit of always-on agents against traditional cron jobs or serverless functions for your specific use cases.
• Stay updated on security best practices for autonomous agents, as documentation from providers like xAI warns against treating them as strict security boundaries.
• Experiment with small, low-risk tasks first to build confidence in the agent’s ability to handle errors and edge cases.