Claude Just Became an Autonomous Workforce
In March 2026, Anthropic shipped a series of updates to Claude Code that fundamentally changed what it can do. This wasn't a minor version bump with bug fixes and polish. This was Claude evolving from an AI assistant that helps you write code into an autonomous agent platform that can control your computer, run tasks on a schedule, accept instructions from your phone, and install new capabilities from a marketplace.
If you're building AI automation for your business — or even just thinking about it — these changes matter. Here's what dropped, what each feature actually does, and what it signals about where AI agents are heading.
Computer Use: Claude Can See and Click Your Screen
This is the feature that gets the most attention, and for good reason. Claude Code can now control your computer — moving the mouse, clicking buttons, scrolling through pages, opening applications, and navigating your desktop just like a human operator would.
Here's how it works in practice. You give Claude a task that requires interacting with a desktop application — something that doesn't have an API or a command-line interface. Instead of telling you "I can't do that," Claude opens the application, navigates the interface, clicks the right buttons, fills in the right fields, and completes the task.
Why This Matters for Businesses
Most business software doesn't have APIs. Your industry-specific CRM, your legacy accounting system, your vendor portal, your insurance quoting tool — these are applications with graphical interfaces designed for humans to click through. Until now, automating these workflows required either expensive custom integrations or robotic process automation (RPA) tools that cost $50,000+ per year.
Computer Use changes the economics completely. Claude can interact with any application that a human can see on screen. No API required. No integration development. No RPA license.
For a plumbing company that uses a specific dispatch software without an API, Claude can now open that software, enter job details, assign techs, and update schedules — all through the graphical interface. For a real estate agent using a proprietary MLS system, Claude can pull comps, enter listing data, and generate reports by navigating the actual application.
The implications for trade businesses and service companies are enormous. Most of the software these businesses use was never built for automation. Computer Use makes it automatable anyway.
The Permissions Model
Anthropic built Computer Use with a deliberate permissions framework. Claude asks for explicit permission before accessing your screen. It requests authorization before interacting with new applications. You can stop it at any point. Certain applications are off-limits by default for security.
This is the right approach. An AI agent that can click anything on your screen needs clear boundaries, and Anthropic is being thoughtful about building those boundaries into the product from day one.
Computer Use is currently in research preview for Claude Pro and Max subscribers, running on macOS. It will expand to other platforms as the technology matures.
Dispatch: Text Your Desktop AI From Your Phone
Dispatch launched on March 17, 2026 as a research preview inside Claude Cowork, and the pitch is simple: text your desktop AI from your phone and come back to finished work.
Here's the workflow. You're away from your desk — at a job site, in a meeting, driving between appointments. You think of something that needs to get done. You pull out your phone, send a message to Claude through Dispatch, and Claude executes the task on your desktop computer. When you get back to your desk, the work is done.
Dispatch pairs your mobile device with your desktop application. Your code and files stay local — only the chat messages are encrypted and transmitted between devices. Claude executes the actual work on your desktop, so nothing sensitive leaves your machine.
Real-World Use Cases
For business owners, Dispatch turns dead time into productive time:
- **From a job site:** "Claude, update the Henderson proposal with the revised materials list I dictated this morning and email the PDF to the client."
- **From your truck:** "Claude, check this morning's leads, respond to the two highest-priority ones with our standard intro, and schedule follow-up calls for tomorrow."
- **From a meeting:** "Claude, pull the Q1 revenue numbers, create a comparison chart vs. Q4, and save it to the reports folder."
- **From the couch on Sunday night:** "Claude, review tomorrow's schedule, prep job folders for each appointment, and send confirmation texts to all customers."
The key insight is that business owners don't stop thinking about their business when they leave their desk. Dispatch means they don't have to wait to act on those thoughts either.
Scheduled Automation
Dispatch isn't just on-demand. It supports cron-style scheduling, which means you can set up recurring automations that run without any manual trigger. Check your inbox every morning at 6 AM and flag urgent items. Pull sales metrics every Monday and format a weekly report. Monitor your Google reviews every evening and alert you to new ones.
This transforms Claude from a tool you use into a system that works for you — even when you're not thinking about it.
The /loop Skill: Cron Jobs for AI
The `/loop` skill is the quickest way to set up a recurring task. You pass an interval and a prompt, and Claude sets up a background cron job that fires automatically.
The syntax is simple: `/loop 5m check deployment status` runs a check every 5 minutes. `/loop 1h review inbox and summarize new messages` runs hourly. `/loop 24h generate daily sales report` runs once a day.
Here's what makes `/loop` smart: recurring tasks automatically expire after 3 days, with one final execution before deletion. This prevents forgotten loops from running indefinitely. You can have up to 50 concurrent tasks per session.
Business Applications
For businesses running AI automation, `/loop` is the scheduling backbone:
- **Monitoring:** Check system health, API uptime, or service status at regular intervals.
- **Reporting:** Generate hourly, daily, or weekly reports automatically.
- **Lead management:** Check for new form submissions or inquiries every 15 minutes and respond immediately.
- **Social media:** Monitor mentions, reviews, or comments on a schedule.
- **Inventory:** Check stock levels against reorder thresholds every few hours.
The 3-day expiration is a thoughtful safety feature. In business automation, runaway processes are a real risk. A loop that checks inventory every hour is useful. That same loop running forgotten for six months, triggering alerts and actions nobody's reviewing, is a liability. The automatic expiration forces intentional renewal.
Voice Mode: Talk to Your Code
The `/voice` command activates push-to-talk interaction with Claude Code. Hold the spacebar, speak your command, and Claude executes it. This works in 20 languages.
For developers, this means hands-free pair programming. For business owners working with Claude to build automations, it means describing what you want in natural language instead of typing detailed instructions.
Combined with Computer Use, Voice Mode creates a genuinely conversational automation experience. You can tell Claude what to do, watch it do it on your screen, and course-correct in real-time through voice commands.
Background Agents: Parallel Execution
Claude Code can now run multiple background agents simultaneously, each working on a separate task in an isolated Git Worktree. This prevents interference between concurrent processes.
In practical terms, you can have Claude working on three things at once — updating a customer report, fixing a bug in your automation workflow, and generating marketing content — without any of those tasks conflicting with each other.
For businesses building AI automation systems, this is significant. It means Claude can develop, test, and refine multiple automation workflows simultaneously, dramatically reducing the time from concept to deployment.
The Plugin Marketplace: One-Click Capabilities
Anthropic launched a plugin marketplace where developers can share and install bundles of MCP servers, skills, and tools. Instead of manually configuring each integration, you can now install a plugin with one click and get a complete capability set.
This is how ecosystems scale. Today, connecting Claude to your CRM, your email system, and your project management tool requires separate configuration for each. Plugins bundle those connections into installable packages. Need a real estate agent toolkit? Install the plugin. Need a trade business operations suite? Install the plugin.
For businesses, the plugin marketplace lowers the barrier to getting value from Claude. Instead of technical setup, you browse a marketplace, install what you need, and start working.
The Model Upgrade: Opus 4.6 and 1M Context
Behind all these features sits an upgraded model — Opus 4.6 — with improved reasoning capabilities. For Max and Enterprise subscribers, the context window expands to 1 million tokens, and the maximum output extends to 128K tokens.
In practical terms, this means Claude can now hold an entire codebase in context, process massive documents, and generate comprehensive outputs without losing track of the details. For business automation, it means Claude can understand your entire operational workflow — all your processes, all your data relationships, all your business rules — and build automation that accounts for the full picture.
What This Signals About the Future of AI Agents
Step back from the individual features and look at the trend. Claude has gone from:
- A text-based assistant to a system that can see and control your screen
- An on-demand tool to a system that runs tasks on a schedule without human trigger
- A desktop-only experience to a system you can command from your phone
- A single-threaded helper to a system that runs multiple parallel agents
- A manually configured tool to a system with a plug-and-play marketplace
This is the trajectory of every technology that becomes infrastructure. It starts as a tool, becomes a platform, and eventually becomes invisible — running in the background, handling work, and only surfacing when it needs human judgment.
For businesses, the message is clear: AI agents are rapidly becoming capable enough to handle real operational work. The companies that start building their automation strategies now — understanding what can be automated, what should stay human, and how to structure the handoffs — will be dramatically ahead of competitors who are still treating AI as a novelty.
How Wolf Intelligence Uses These Tools
At Wolf Intelligence, we build AI automation for businesses — and we use these tools every day to build and deploy our product suite. Claude Code is part of our development infrastructure for building products like AI Auto Attendant, Invoice Chaser, Review Guard, and the rest of the Wolf Pack.
The Computer Use capability is particularly relevant for our clients in trades and services. Many of these businesses run industry-specific software that was never designed for API integration. Computer Use opens the door to automating workflows in those systems without waiting for the software vendor to build an API.
Dispatch and `/loop` are foundational for the always-on automation our clients expect. When we deploy an Invoice Chaser that monitors receivables and sends follow-up reminders, that system needs to run on a schedule without human babysitting. When a client wants to manage their AI automation from the job site, Dispatch makes that possible.
These aren't theoretical use cases. This is how modern AI automation gets built.
Frequently Asked Questions
**Is Claude Code Computer Use available on Windows?** Not yet. Computer Use is currently in research preview and available only on macOS for Claude Pro and Max subscribers. Anthropic has indicated plans to expand platform support as the feature matures.
**Does Dispatch send my files or code to the cloud?** No. Dispatch only transmits encrypted chat messages between your phone and desktop. All code, files, and actual task execution happen locally on your desktop. Nothing sensitive leaves your machine.
**Can /loop tasks run indefinitely?** No. Recurring tasks created with `/loop` automatically expire after 3 days, with one final execution before deletion. This is a deliberate safety feature to prevent forgotten automations from running unchecked. You can renew tasks that need to continue.
**What does the plugin marketplace cost?** The plugin marketplace is part of Claude Code. Individual plugins may be free or paid depending on the developer. The marketplace itself doesn't add cost beyond your existing Claude subscription.
**How does Computer Use handle security and permissions?** Claude requests explicit permission before accessing your screen and before interacting with new applications. Users can stop Claude at any point during computer use. Certain applications are restricted by default, and Anthropic includes automatic scanning for prompt injection attacks. The system is designed with a permission-first approach where the user stays in control.
Take the Next Step
Want to understand where AI automation can eliminate manual work in your business? Wolf Intelligence's free AI Readiness Assessment identifies your highest-ROI automation opportunities in 10 minutes.
[Take the free AI Readiness Assessment](/ai-readiness) and see what's possible with today's AI tools.