Anthropic’s Claude Code and Cowork can control your computer

Anthropic updated Claude to autonomously perform tasks on users' computers within its Code and Cowork AI tools, available as a research preview for Claude Pro and Max subscribers. The capability is currently limited to macOS, requires the Claude desktop app paired with the mobile app, and prefers connectors (Slack, Google Workspace) but can fall back to controlling the browser, mouse, keyboard and display with explicit user permission. Anthropic notes the feature pairs well with its Dispatch cross-device workflow but can be slower than direct integrations and may require retries for complex tasks.

Analysis

This feature is an acceleration of a broader bifurcation: lightweight agent capabilities that orchestrate user devices vs. heavy cloud inference. In the near term (weeks–months) adoption will be supply-constrained by client-platform integration and enterprise security review cycles, so revenue lift to infrastructure players will be lumpy rather than immediate. Over 6–24 months, however, enterprises will prefer either first-party secure connectors or on-prem inference to avoid clicking-through attack surfaces, which increases demand for datacenter GPUs, private model hosting, and secure connector tooling.

Competitive dynamics favor companies that can offer end-to-end secure stacks: firms that combine model hosting, enterprise connectors, and hardware provisioning will capture disproportionate wallet share. Public hyperscalers with integrated workspace suites face margin pressure from third-party agents that bypass APIs by automating UIs, forcing them either to harden connectors (raising costs) or to monetize agent integrations. Expect increased sales cycles for Workspace/Collaboration products as IT teams demand auditability and access controls — a headwind for low-touch SaaS adoption but a tailwind for paid enterprise tiers.

The key risk is a security incident that maps directly to automated control of endpoints; one high-profile breach could trigger regulatory scrutiny and an enterprise pause lasting quarters. A faster reversing catalyst would be cross-platform parity (Windows/Android) or improved native integrations from incumbents that make third-party agent routes redundant. Monitor early bug reports and SOC/ISO defensive guidance: those will be leading indicators that adoption is either scaling or stalling.

Key Decisions for Investors

• Long NVDA via a defined-risk call spread (6-months): buy-to-open NVDA 6m OTM call / sell higher strike call to fund ~1% NAV exposure. Rationale: captures incremental server/GPU demand from enterprises hedging to on-prem/private inference; target 2x–3x payoff if NVDA rerates on sustained enterprise GPU uptake; max loss = premium paid.
• Pair trade — long NVDA (1% NAV) / hedge with short GOOGL 3-month 10% OTM put spread (0.25% NAV bought protection): asymmetric bet that hardware demand outpaces monetization wins for workspace vendors. This limits downside from a sudden privacy/regulatory pullback that would hit integrators and search-advertising sensitivity.
• Small, long-term LEAP on GOOGL (12–24 months): buy modest exposure to capture larger enterprise wallet-share shift into hosted models and workspace monetization—size <=1% NAV. Use this as a conviction trade only if you see IT procurement wins or GA connector partnerships; otherwise keep position light for regulatory tail risk.