Enough of clicking through the same menus, filling the same forms, and manually executing the same workflows you've done a thousand times. On March 23, 2026, Anthropic flipped a switch that turned Claude from an AI assistant you talk to into an AI agent that works for you — autonomously controlling your Mac, navigating your apps, and completing tasks while you focus on what actually matters. And the reason this terrifies some people and thrills others comes down to one question nobody in your industry is asking loudly enough: what exactly does your job look like when AI can do it while you sleep?
You were told AI would eventually replace certain jobs. You were not told it would do it by literally sitting at your computer and using it the way you do — clicking, typing, navigating, filling forms, running commands, and completing entire workflows without a single human action. That day arrived. Anthropic's launch tweet got 130,000 likes and 70 million views in 24 hours. The feature was described, succinctly, by one developer on social media as: "that's not an AI assistant anymore, that's infrastructure."
What Is Claude Code Computer Use — and What Can It Actually Do?
On March 23, 2026, Anthropic launched computer use capabilities for Claude Code and Claude Cowork on macOS. This is not a chatbot improvement. It is a fundamental change in what AI agents can do. Previously, Claude could help you write code, answer questions, draft documents, and analyze data — but it required you to execute every action. You were the hands. Claude was the brain that instructed you. Computer use reverses that relationship. Now, Claude Code can see your screen, move your cursor, click buttons, type into any application, navigate your browser, open files, fill spreadsheets, run terminal commands, and complete multi-step workflows across multiple applications — autonomously, without your intervention, while you're away from your desk.
The official Anthropic demo showed a user asking Claude to export a pitch deck as a PDF and attach it to a calendar meeting invite — all while the user was away from their Mac. Claude opened the application, exported the file, navigated to the calendar, located the meeting, attached the PDF, and confirmed — with zero human clicks. That's not a demo. That's your assistant.
How It Works — The Architecture Behind Autonomous Computer Control
Claude Code's computer use capability operates through a layered approach that prioritizes efficiency and security. When you assign a task, Claude first checks whether a direct connector exists for the relevant application. Anthropic has already built connectors for Slack, Gmail, Google Drive, Google Calendar, Notion, and other major productivity tools. When a connector is available, Claude uses it directly — fast, reliable, and without screen interaction.
When no connector exists — for legacy software, niche tools, or applications with no API — Claude falls back to screen control. It takes a screenshot of your display, analyzes what it sees using its vision capabilities, determines the correct action, and executes it: moving the cursor, clicking the right element, typing the required input. Then it screenshots again to verify the result and continues. This loop — screenshot, analyze, act, verify — is the same process a human would use to operate an unfamiliar application. Claude just does it faster and doesn't need breaks. The context window has been expanded to 1 million tokens specifically to support these long, complex workflows. The maximum output has been doubled to 128,000 tokens. A new safety classifier handles autonomous decision-making for low-risk actions, reducing the need for constant human approval.
The Dispatch Feature — Your Mac Works While You Sleep
Paired with computer use is Dispatch — a feature that lets you assign Claude tasks from your iPhone and return to finished work on your desktop. Leave your Mac running. Open the Claude app on your phone. Tell Claude to research a topic and compile a report, find and organize files matching certain criteria, fill out a series of forms across multiple web applications, run a test suite and summarize the results, or schedule and send a series of emails. Claude executes. You receive a push notification when it's done — or when it needs your approval for something it's uncertain about. One developer described the compound effect: "Combine this with /schedule and you've basically got a background worker that can interact with any app on a cron job. That's not an AI assistant anymore, that's infrastructure."
The Plugin Architecture — When Claude Becomes a Domain Expert
This is where Anthropic's enterprise ambitions become unmistakably clear. Anthropic is building a plugin ecosystem that bundles skills, connectors, and sub-agents into single installations that turn Claude into domain specialists. The company already lists plugins for legal workflows (contract review, NDA triage), finance (journal entries, reconciliation, variance analysis), and brand voice management (analyzing documents to enforce brand guidelines automatically). An AI Automation Engineer at one enterprise client described connecting Cowork to her company's entire tech stack and asking it to identify engineering bottlenecks — across code repositories, ticketing systems, communication channels, and deployment pipelines — simultaneously. This is not a tool anymore. This is infrastructure.
The Security Questions Nobody Is Answering Satisfactorily
Here is the part that needs honest discussion. When Claude uses your computer, it takes screenshots of your screen continuously. Any information visible during the session — emails, documents, financial data, private messages — is captured and processed. Anthropic's documentation recommends closing sensitive files before starting a session and denying permissions to banking or medical applications. Security researchers have already documented what they call "ZombAI" attacks: malicious web pages that embed hidden instructions which Claude reads and executes without the user's knowledge. In one scenario, a compromised web page could theoretically turn Claude into a command-and-control node for a remote attacker. Anthropic has built multiple guardrails: per-app permissions, sensitive app blocking by default, and explicit authorization for permanent deletions. The company labels this a "research preview" specifically because the security model is still maturing. The question isn't whether computer use is powerful. It clearly is. The question is whether the guardrails are moving as fast as the capabilities.
What This Means for the Future of Work
Anthropic's own economic index, published in March 2026, tracks how AI is reshaping labor markets across sectors. The data shows AI adoption is accelerating unevenly — with knowledge workers in technology, finance, and professional services seeing the most dramatic shifts. Marketing teams that previously required five people to manage campaigns, analytics, content production, and reporting are now discovering that one person with the right AI agent stack can do the work of the entire team. Operations workflows that required dedicated headcount to navigate legacy systems with no APIs are being automated by computer use agents that don't need APIs — they just use the screen. The question is not whether this changes your industry. It already has. The question is whether you are the person deploying Claude to do the work — or the person being replaced because someone else did.
Want to follow every major AI move the moment it happens? Join thousands of readers at fabioraphaelli.com and get the biggest AI news every week.
Frequently Asked Questions
What is Claude Code Computer Use?
Claude Code Computer Use is a feature launched on March 23, 2026, that allows Claude to autonomously control a Mac — clicking, typing, navigating apps, running commands, and completing workflows without human intervention. It is available as a research preview for Claude Pro and Max subscribers on macOS.
What can Claude Code do on my Mac?
Claude Code can open applications, fill forms, navigate web browsers, run terminal commands, manage files, export documents, edit spreadsheets, interact with desktop software that has no API, and complete multi-step workflows across multiple apps — all autonomously. It uses direct connectors for apps like Gmail and Slack when available, and falls back to screen control otherwise.
What is the Dispatch feature in Claude Code?
Dispatch allows you to assign Claude tasks from your iPhone and return to completed work on your desktop. You leave your Mac running, send Claude a task from your phone, and receive a push notification when the work is finished or when Claude needs approval to proceed.
How does Claude Code handle security and privacy?
Anthropic has built guardrails including per-app permissions, sensitive app blocking, and authorization requirements for irreversible actions. However, it's labeled a research preview, and Anthropic recommends closing sensitive files during sessions. Security researchers have also identified potential vulnerabilities like "ZombAI" attacks, where malicious web pages could embed hidden instructions.
Will Claude Code Computer Use replace jobs?
Anthropic's economic index suggests significant AI-driven workforce shifts, particularly in knowledge work. While it may not replace entire jobs immediately, it will automate specific tasks and workflows, especially those involving GUI-based applications. This creates a shift where individuals who can orchestrate AI agents will be highly valued.
What is the plugin architecture for Claude Code?
Anthropic is developing a plugin ecosystem that bundles specialized skills, connectors, and sub-agents. These plugins turn Claude into domain experts for specific tasks like legal workflows, financial analysis, or brand voice management, allowing it to integrate deeply with enterprise tech stacks.









