Claude Code and Cowork can now use your computer

Anthropic has enabled Claude Code and Claude Cowork to control users' computers in a research preview for Claude Pro and Claude Max subscribers on macOS, allowing the AI to open files, use the browser and run developer tools. The agent will prioritize connectors to services like Google Workspace and Slack but can execute tasks without them and should ask for permission; Anthropic cautions against using the feature for sensitive data and will iterate based on user feedback. The update also integrates with Anthropic's Dispatch and extends the January-launched Claude Cowork aimed at casual programmer users.

Analysis

The emergence of agents that can act on end-user endpoints materially shifts the battleground from cloud-model quality to control of connectors, identity and orchestration layers. Firms that own enterprise identity, MDM, and first-party collaboration stacks stand to either capture a new monetizable surface (connector fees, premium integrations) or face margin pressure as independent agents aggregate productivity value away from core apps. Expect a multi-quarter chase as incumbents rush to productize safe connectors and as integrators build certified “agent rails” that become de facto standards.

Security and regulatory risk are the primary brake on adoption. A single high-profile leakage or improper action by an agent would trigger rapid tightening: emergency guidance from regulators, enterprise-wide disablement, and insurance re-pricing — a timetable measurable in days to weeks after an incident and with earnings impact visible in the next quarter. Conversely, credible third-party attestations (SOC2/ISO+attestation for agent behavior) or major platform partners committing to certified connectors will meaningfully de-risk adoption over 6–18 months.

For investors, the trade is between near-term headline risk and longer-run structural monetization of workflow automation. Catalyst windows are clear: productization announcements by major cloud/workspace vendors and any publicized security incidents. Monitor connector certification programs, pricing changes for workspace APIs, and regulatory dialog; these will be the quickest leading indicators of winners versus those forced into costly remediation and feature rollbacks.

Key Decisions for Investors

• Directional growth-tech play: Buy GOOG 6–12 month call spreads to express upside from Google monetizing agent-driven workflows and defending search/workspace moats; target 2–3x upside if integrations broaden, cap risk to premium paid with a 20–30% downside sensitivity if adoption is delayed.
• Security re-rate trade: Go long CRWD (or buy 3–9 month call options) to capture demand for endpoint protection and behavioral monitoring if agents scale; set a stop-loss at 15% and a target of 40–60% on a successful re-rate into security as enterprises increase budgets.
• Event hedge / asymmetric downside: Buy 3-month puts on GOOG sized to cover 25–50% of your directional exposure to protect against a regulatory/security shock that would force mass disablement of device agents; treat this as insurance—cost is certainty vs open downside risk.
• Relative idea: Long CRWD / short GOOG (or GOOGL) over 3–9 months to play a rotation from ad/search-levered incumbents into security infrastructure; aim for a 1.5:1 expected return profile with explicit event triggers (public breach, connector certification rollouts) to rebalance.