Salesforce just watched $15 billion in market cap vanish in a single afternoon, and the culprit wasn’t a bad earnings report—it was a demo of a computer mouse moving by itself. When Anthropic unveiled Claude’s new “computer use” capability, it didn’t just update a chatbot; it fired a shot across the bow of every subscription software company on the planet. If an AI can navigate a desktop, fill out spreadsheets, and click “submit” in a browser just like a human, the “Software as a Service” era is officially entering its most volatile chapter yet.
| Attribute | Details |
| :— | :— |
| Difficulty | Intermediate (Requires API knowledge for full automation) |
| Time Required | 15–30 minutes for initial setup & testing |
| Tools Needed | Claude Pro/Max, Anthropic Dispatch, Desktop environment |
The Why: The Death of the “Integration Middleman”
For a decade, Salesforce and its peers thrived because they were the “system of record.” You paid for the seat because you needed a human to input data, move leads, and manage workflows. Integrating these systems was a nightmare of expensive APIs and “glue” code.
Anthropic’s new model solves this by bypassing the API entirely. By “seeing” the screen and “using” the keyboard, Claude can bridge the gap between a PDF pitch deck and a CRM entry without a single line of integration code. Investors are spooked because if the AI can do the work of a sales op specialist directly on the screen, the premium pricing of complex SaaS platforms becomes much harder to justify. This isn’t just a new feature; it’s the beginning of the “Agentic UI.” This shift highlights why Wall Street Just Realized AI Is Coming for the Middlemen, as traditional software moats begin to evaporate.
Step-by-Step: Putting Claude to Work on Your Desktop
This feature is currently in research preview for Pro and Max subscribers. Here is how to move from “chatting” to “executing.”
- Request Access: Navigate to your Claude dashboard and enable the “Computer Use” preview features. Anthropic’s Claude 3.5 Sonnet Just Got a Massive Brain Transplant—and It Can Now Use Your Mouse, making it the first major LLM to interact with a PC like a human.
- Sync with Dispatch: Install the Dispatch app on your smartphone. This acts as your remote control, allowing you to send “missions” to your desktop while you’re away.
- Define the Sandbox: Run Claude in a dedicated environment or a secondary browser. Until safeguards are perfect, you don’t want an AI clicking buttons in your primary banking tab.
- Issue the Mission: Use an action-oriented prompt. Instead of “Summarize this,” try: “Open my Downloads folder, find the latest CSV, upload the data into the Salesforce ‘New Leads’ tab, and send a Slack confirmation to the team.” These specialized AI agents are designed to handle multi-step tasks that previously required manual oversight.
- Monitor the Iteration: Claude will take screenshots of your screen to “see” where the buttons are. Watch the live feed to ensure it doesn’t get stuck in a login loop.
💡 Pro-Tip: Claude prioritizes “precise tools” (like direct Slack connectors) before it resorts to moving your mouse. To save on tokens and increase speed, always check if a native connector exists in your Claude settings before asking it to manually click through a website.
The Buyer’s Perspective: Anthropic vs. The Field
While Microsoft’s “Recall” faced a PR disaster over privacy, Anthropic is taking a “permission-first” approach. Every time Claude wants to touch a new application, it asks for a green light.
Compared to Salesforce’s own Agentforce, Anthropic has a distinct advantage: Generality. Agentforce is brilliant within the Salesforce ecosystem, but it struggles once you leave the “Blue Cloud.” Claude doesn’t care if you’re using a legacy accounting tool from 2004 or a modern web app; if there are pixels on the screen, Claude can interact with them. This is part of a larger Claude ecosystem designed to prioritize high-performance productivity over walled-garden limitations.
However, Salesforce has the “Trust Layer.” Enterprises might be hesitant to let a third-party LLM “see” their entire screen, whereas they already trust Salesforce with their most sensitive customer data. For now, Anthropic is the superior “doer,” but Salesforce remains the superior “vault.” This battle for enterprise control is why many argue that Claude Enterprise is becoming the new secure operating system for modern business data.
FAQ
Q: Will Claude delete my files by accident?
A: It’s possible but unlikely. Claude requires explicit permission to access new apps, and you can kill the process instantly. However, Anthropic explicitly warns against using it with sensitive data during this research phase.
Q: Do I need to know how to code to use this?
A: No. Unlike traditional RPA (Robotic Process Automation) tools like UiPath, you don’t need to build “flows.” You just tell it what you want in plain English, and the model figures out the clicks.
Q: Why did Salesforce stock specifically drop?
A: Investors fear “SaaS Cannibalization.” If an AI agent can handle sales, business development, and data entry, companies may eventually need fewer “seats” of expensive software, hitting the bottom line of CRM giants.
Ethical Note/Limitation
Despite its “vision,” Claude still struggles with complex UI elements like dropdown menus that disappear on hover or high-latency remote desktops where the screen lag confuses its “eyes.”