Claude can now control macOS computers directly, opening apps, navigating browsers, and clicking buttons like a human operator.
Computer Use pairs with Dispatch mobile tool, letting you assign tasks from phone while Claude completes them remotely on your Mac.
Claude Code Channels enable async coding via Telegram or Discord, letting developers ping background sessions and receive task results on mobile.
Feature works with any application lacking APIs through screen control, making it useful for structured workflows across fragmented tool ecosystems.
Research preview shows impressive results for forms and navigation but remains less reliable for creative or ambiguous tasks.
Claude Just Had Its Biggest Week. Here's What Actually Matters.
I've been building with Claude Code every single day for the past month. RAXXO Studios runs on it - from generating content to managing Shopify automation to writing the code that powers my SaaS app. So when Anthropic drops a dozen major updates in a single week, I pay attention. Not as a hype reporter, but as someone whose entire workflow depends on this tool.
Here's what landed in March 2026, what it actually means for solo creators and developers, and where I think this is all heading.
Computer Use: Claude Can Now Operate Your Mac
This is the headline feature, and for good reason. As of March 23, Claude can directly control your macOS computer. It opens apps, navigates your browser, fills in spreadsheets, clicks buttons - basically anything you'd do with a mouse and keyboard.
The way it works is smart. When you give Claude a task, it first checks if it has a direct integration (like Google Calendar or Slack). If it doesn't, it falls back to actually controlling your screen like a human would. That fallback approach means it can work with virtually any application, even ones that have no API.
Who gets it: Claude Pro and Max subscribers on macOS. Windows support is coming later. It pairs with Dispatch, a mobile companion tool that lets you assign tasks from your phone and have Claude complete them on your computer while you're away.
My take: This is a research preview, and Anthropic is upfront that it's "still early compared to Claude's ability to code or interact with text." I've tested it, and it's impressive for structured workflows - filling forms, navigating familiar UIs, copying data between apps. It's less reliable for creative or ambiguous tasks. But the trajectory is clear: Claude is becoming an actual computer operator, not just a chat assistant. For solo creators juggling a dozen tools, that's a genuine time multiplier.
Anthropic says they've built safeguards in - Claude asks permission before accessing new apps. That's the right approach for something this powerful.
Claude Code Channels: Async Coding via Telegram and Discord
Released March 20, Claude Code Channels let you control a running Claude Code session through Telegram or Discord. Your session sits in the background on your machine, and when you ping it from your phone, it wakes up, runs the task, and messages you back when it's done.
This is a plugin-based architecture (requires Claude Code v2.1.80+ and the Bun runtime), and the two launch platforms are Telegram and Discord. The community is already requesting Slack, WhatsApp, and iMessage support.
My take: This changes the relationship with your dev environment. I can be on the train, message Claude via Telegram to refactor a component or run an audit script, and get the results pushed back to me. No laptop required. No SSH session to maintain. The session has full filesystem, MCP, and git access - it's your actual development environment, just remotely accessible.
For anyone running async workflows - deploying, testing, content generation - this is a big deal. Fire off five tasks in the morning, check the results over lunch. That's a real productivity pattern, not a demo.
Voice Mode: Talk to Your Code
Voice mode shipped on March 3 with the /voice command. It uses push-to-talk (hold spacebar, speak, release) rather than always-on listening, which is the right call for a development tool. You don't want your AI coder accidentally picking up a phone conversation.
In March, Anthropic added 10 new languages (Russian, Polish, Turkish, Dutch, Ukrainian, Greek, Czech, Danish, Swedish, Norwegian), bringing the total to 20. They also optimized transcription for technical terms and repository names - crucial when you're talking about getUserSession or trackUsage and need it transcribed correctly.
My take: I was skeptical about voice for coding. Then I tried it for architecture discussions - talking through a refactor while pacing around my studio. It's surprisingly natural for high-level decisions. For actual code edits, I still prefer typing. But for "hey, look at this component and tell me what's wrong" or "explain how this API route handles authentication" - voice is faster than typing a paragraph of context.
It's included at no extra cost for all paid tiers. That's the right pricing decision.
Opus 4.6 and the 1M Context Window
On March 13, Anthropic made the 1 million token context window generally available for both Opus 4.6 and Sonnet 4.6 - at standard pricing. No premium tier, no "long context surcharge." A 900K-token request costs the same per-token rate as a 9K one.
The numbers: up to 128K output tokens, up to 600 images or PDF pages per request (6x the previous limit of 100), and Opus 4.6 hits 78.3% on the MRCR v2 benchmark at 1M tokens - the highest recall rate among frontier models.
API pricing: Opus stays at 5/25 USD per million tokens (input/output), Sonnet at 3/15 USD.
My take: The context window expansion is one of those upgrades that sounds incremental but fundamentally changes what's possible. I can now feed Claude an entire codebase - not snippets, the whole thing - and get coherent, context-aware responses. For RAXXO Studios, that means Claude understands the full project structure when I ask it to add a feature or debug an issue. No more "here's the relevant file" gymnastics.
The 128K output limit is equally important. Claude can now generate entire features, complete documentation, or comprehensive analysis in a single response without truncation.
/effort and Ultrathink: Control How Hard Claude Thinks
The /effort command (v2.1.68, March 4) gives you three persistent effort levels: low, medium, and high. Set it once and it stays until you change it. Low is fast and cheap for simple tasks. High uses the full 31,999-token thinking budget for complex architecture decisions.
Then there's ultrathink - a power word you can add to any prompt to trigger maximum reasoning depth. It was originally in Claude Code, got removed, and after widespread community feedback about quality degradation, Anthropic brought it back. Ultrathink maps to the high effort level. The absolute maximum (no token constraints) is only accessible via the /effort setting or the API.
My take: This is about cost and speed control. Quick formatting fix? Low effort, instant response. Redesigning a database schema? High effort, let it think. I use low for 70% of my interactions and switch to high when I'm stuck on something complex. It's the kind of power-user feature that saves real money over time if you're on metered API usage.
/loop: Recurring Tasks on Autopilot
The /loop command (v2.1.76, March 17) lets you set up recurring tasks that Claude runs automatically. I already wrote a deep dive on /loop, so I'll keep this brief: it's the foundation for Claude as a background worker, not just a chat partner.
Combined with Channels, you can set up loops that run on your machine and report results to your phone via Telegram. That's a monitoring system built from a chat interface.
100M USD Partner Network
On March 12, Anthropic announced the Claude Partner Network with a 100M USD investment. Anchor partners include Accenture (training 30,000 professionals on Claude), Deloitte, Cognizant, and Infosys. The network offers training, technical support, joint go-to-market investment, and a new technical certification - all free to join.
My take: This matters for solo creators and small studios more than you'd think. When enterprise consulting firms train tens of thousands of people on Claude, it creates an ecosystem. More integrations, more MCP servers, more tools that work with the platform you're already building on. Rising tide.
Office Integration: Excel and PowerPoint
The March 11 update gave Claude shared context across Excel and PowerPoint. Actions in one application are now informed by what's happening in the other. Anthropic also shipped pre-built skills: formula auditing and balance-sheet integrity checks for Excel, competitive landscape deck generation for PowerPoint.
My take: I don't live in Excel or PowerPoint, but many of my consulting clients do. The ability to say "audit this financial model for errors and then build a summary deck from it" in one conversation - with Claude understanding both files - is genuinely useful for business workflows. It's Claude becoming an actual office assistant, not just a coding tool.
What I'm Most Excited About
If I had to pick one thing from this entire update cycle, it's the combination of Channels + Computer Use + /loop. Each feature is useful on its own. Together, they create something new: an AI agent that lives on your machine, can operate any application, runs tasks on a schedule, and reports back to your phone.
I'm already sketching workflows: automated content audits that run nightly and ping me with issues, Shopify inventory checks that flag problems before they become customer complaints, design asset generation that queues up while I sleep.
The shift is clear: from "Claude as a chat partner" to "Claude as an autonomous team member." For solo creators and small studios, that's not incremental. That's a category shift. The tools are in research preview, they'll get more reliable, and the workflows people build on top of them will be the real story of 2026.
I'll be documenting everything I build with these new features right here on the Lab blog. If you're experimenting with Claude Code too, I'd love to hear what you're building - drop me a line at help@raxxo.shop.
This article contains affiliate links. If you sign up through them, I earn a small commission at no extra cost to you.
This article contains affiliate links. If you sign up through them, I may earn a small commission at no extra cost to you. (Ad)
Top comments (0)