‘Claude, Resize These Photos’ – Anthropic’s Agentic AI Will Run Photoshop For You

Anthropic has introduced agentic capabilities in Claude that let the AI operate a user's macOS computer to perform tasks like batch-resizing photos and adding logos; the feature is a research preview in Claude Cowork and Claude Code and is available to Claude Pro and Claude Max subscribers on macOS. The company plans to prioritize connectors to services like Google Workspace and Slack, but the agent can also operate apps directly when connectors aren't available. CNET-cited experts and Anthropic warn of security and unpredictability risks—users are advised not to grant access to sensitive files and to treat this as an experimental feature.

Analysis

Agentic UI automation converts human GUI interactions into a programmable workflow layer, which will reallocate IT spend from bespoke RPA projects to connector/orchestration platforms; expect enterprises to shift 5–10% of low-skill desk work to these agents within 12–24 months, concentrating demand on identity, endpoint telemetry, and logging infrastructure. That shift is non-linear: a 10% automation penetration can create a 20–40% uplift in telemetry volume per seat (more automation = more noisy event streams), increasing SOC headcount and SaaS security spend disproportionately to the automation savings.

Strategically, cloud-platforms that own connectors and identity (e.g., office suites, IaaS) gain marginal lock-in and a new usage metric—agent API calls—that can be monetized; vendors that are downstream (standalone creative apps) face bifurcated outcomes where pros keep paying for capabilities but SMB/freelancer usage and per-image revenue are fungible and at risk. Hardware demand is asymmetric: local, low-latency inference and secure enclaves matter for high-trust use cases, but widespread adoption hinges on enterprise security assurances—one high-profile breach or regulation could pause deployments within weeks and slow revenue realization for a year.

The market is underpricing enterprise security upside and over-indexing on near-term creative-app disruption. The dominant path for value capture is via security/identity SaaS and cloud connectors, not consumer creative tooling, so exposure should favor companies with sticky telemetry ingestion and enterprise contracts. Watch three short-duration catalysts: first large enterprise procurement announcements (3–9 months), a public security incident (days–weeks), and connector partnerships with major cloud/office vendors (1–6 months) to re-rate winners and losers.

Key Decisions for Investors

• Long Palo Alto Networks (PANW) — buy shares or 6–12 month call spreads. Rationale: direct beneficiary of increased telemetry/SOC demand; target +30% upside if enterprise automation adoption accelerates, capped downside ~15% if adoption stalls. Enter on pullback or after a security contract announcement.
• Long CrowdStrike (CRWD) — purchase shares or Jan 2027 LEAP calls (vertical call spread). Rationale: cloud-native endpoint telemetry scales with agent usage; expect 25–40% upside over 9–18 months as ARR expands, downside 20–30% on macro tech correction. Layer in 25–50% of position on any weakness after a competitor earnings miss.
• Pair trade: Long Microsoft (MSFT) / Short Adobe (ADBE) for 6–12 months (equal notional). Rationale: MSFT captures connector lock-in and enterprise suites; Adobe faces secular pressure in low-skill segments. Target 15–25% relative alpha; risk that Adobe accelerates product-level AI features, which would reverse within quarters.
• Selective long NVIDIA (NVDA) for 12 months (buy-dated calls with downside hedge). Rationale: edge and server inference demand rises as agent workloads scale; expect 15–30% upside if cross-platform expansion occurs, but hedge against valuation and cloud-only model risks which could push a 25–40% drawdown.