Claude can now remotely control your computer, and it looks absolutely wild [Video]

Anthropic added a remote-control capability to Claude Cowork and Claude Code that lets the AI control a user's macOS computer from a paired phone; the feature is restricted to Claude Pro ($20/month) and Max ($100/month) plans and requires the desktop app to be awake. Claude will attempt service connectors (e.g., Slack, Google Calendar) first, then can control browser/mouse/keyboard with explicit user permission; Anthropic advises against granting access to sensitive financial, legal, or medical apps and says Claude is trained to avoid trading, inputting sensitive data, or scraping facial images. Demonstrated use cases include exporting PDFs to calendar invites, starting a dev server and capturing screenshots, and batch photo edits with watermarks.

Analysis

This feature flips a common assumption: the marginal value of connectors falls if a model can reliably operate a user’s UI. That compresses the revenue runway for niche integration builders and traditional screen-scraping tools, while concentrating new incremental spend into endpoint security, identity controls, and telemetry to monitor autonomous UI actors. Expect an initial adoption pattern where prosumers and developer teams act as early signal-sets, followed by cautious enterprise pilots once centralized policy & audit capabilities are mature.

Second-order infrastructure winners are identity (session management, MFA), EDR/SIEM, and secure automation orchestration — not generic cloud compute. If enterprises add even a modest 5–10% of security budgets to cover autonomous agent risk over 12–24 months, that translates into double-digit revenue tailwinds for leading EDR/identity vendors relative to peers. Conversely, incumbents that monetize by selling connectors or RPA licensing face a true substitution risk: UI-driven agents remove many integration maintenance costs and may compress SaaS margins in the integration layer.

Main downside catalysts are a single high-profile misuse or breach that triggers rapid enterprise bans and regulatory scrutiny; such an event can crystallize policy headwinds within weeks and stall growth for quarters. Operational reality also matters: usability, error rates, and auditability will determine whether CIOs allow autonomous UI control; those metrics — not headlines — will drive vendor RFPs over the next 3–18 months.

Key Decisions for Investors

• Long CRWD (CrowdStrike) 6–18 months: allocate 3–5% position. Rationale: incremental enterprise spend on endpoint telemetry and autonomous-agent detection should lift EDR contract sizes. Target +25–40% upside; stop-loss 12%.
• Long OKTA 3–12 months: buy shares or 9–12 month call spread targeting tighter enterprise session controls and MFA monetization. Expect 15–30% upside if identity vendors win expanded auth workflows; downside is slow procurement — cap loss at 15%.
• Long PANW (Palo Alto Networks) 3–12 months: add to security exposure via shares or 6–12 month call options. Network policy and SASE bundles can be upsold to control agent egress and auditing; target 20%+ return, stop 10–12%.
• Short PATH (UiPath) via 9–12 month puts (small size): thematic hedge against RPA feature compression from agent-driven UI automation. If adoption accelerates, PATH could see durable growth deceleration — aim for asymmetric payoff (target 30–50% move) with limited notional exposure.