Anthropic just shipped two features in the same week that change the same thing: how you reach Claude when you are not at your desk.
Claude Code Channels lets developers message Claude Code directly over Telegram and Discord. You text it a task. It writes code in your dev environment. You go back to your life.
Dispatch does the same thing for Cowork — the non-technical side of Claude. Scan a QR code on your desktop, pair your phone, and you can assign tasks to Claude from anywhere. It pulls together reports, handles Slack, prepares presentations, manages files. All running locally on your machine while you are somewhere else.
Two products. Same thesis: the agent should be working even when you are not watching.
What It Is
When you launch Claude Code with the new --channels flag, you are not just opening a chat. You are spinning up a polling service. An MCP server acts as a two-way bridge — incoming messages from Telegram or Discord are injected directly into your active Claude Code session. Claude executes the work and responds through the external platform.
Setup is straightforward. Create a Telegram bot via BotFather, link it to Claude Code with a /telegram:configure command, pair your account with a security code, and your phone becomes a remote control for your development environment. Discord follows the same pattern through the Developer Portal.
It requires version 2.1.80 or higher and the Bun JavaScript runtime. It is currently a research preview.
The OpenClaw Problem
This is a direct shot at OpenClaw — the open-source AI assistant that Peter Steinberger launched in November 2025. OpenClaw lets you talk to AI agents over iMessage, Slack, Telegram, WhatsApp, and Discord. It has 325,000 GitHub stars and a passionate community. It proved the demand was real.
It also proved the security model was fragile. Earlier this year, researchers found a critical vulnerability in OpenClaw with a CVSS score of 8.8. Over 50,000 instances were exposed. When your AI agent has access to your development environment and you are reaching it over consumer messaging platforms, the attack surface is not theoretical.
Claude Code Channels runs in a sandbox. Permissions are explicit. Actions require approval. Enterprise admins control what is enabled. Anthropic is not just matching OpenClaw's functionality — they are wrapping it in the trust infrastructure that enterprise customers demand.
AI YouTuber Matthew Berman summarized it plainly: "They've BUILT OpenClaw."
The Shift
The real story is not the messaging integration. It is the model change.
Until now, working with Claude Code meant sitting at your terminal or your IDE. You typed. Claude responded. You reviewed. You approved. Synchronous. Present. Hands on the keyboard.
Channels moves this to an asynchronous, autonomous partnership. You text Claude a task while walking the dog. Claude works on it. You check in later. The agent is not waiting for you — it is working for you while you are somewhere else.
This is what I have been writing about since Meet My New AI Colleagues. The agentic tools are compressing not just the work but the attention required to direct it. A solo operator does not need to be sitting at a desk to have code being written. The bottleneck was never the AI's ability to write code. It was the human's availability to supervise it.
Channels loosens that constraint. Not all the way — you still need to review and approve. But the gap between "I had an idea" and "the code is written" just got a lot shorter for anyone who is not chained to their laptop.
Built on MCP
This is also a showcase for the Model Context Protocol that Anthropic released as an open standard back in 2024. MCP is the universal connector — the USB-C port for AI. In the Channels architecture, it is the bridge between your messaging app and your code environment.
Every time Anthropic ships a feature built on MCP, the protocol gets more entrenched. Every MCP server someone builds is another reason the ecosystem gravitates toward it. This is platform strategy disguised as developer tooling. And it is working.
Dispatch: The Other Remote Control
Channels is the developer story. Dispatch is the everyone-else story.
Dispatch is a research preview within Cowork — Anthropic's non-coding agent that handles knowledge work, file management, email, and desktop automation. The setup is almost comically simple: open the Claude desktop app, scan a QR code with your phone, and the two devices are paired. Your phone becomes a remote control for whatever Claude is doing on your desktop.
From your phone, you can assign tasks — pull together a report from local files, draft responses in Slack, prepare a presentation from Google Drive, organize a folder. Claude executes everything locally on your machine. Your files never leave your computer. Every action still requires confirmation through the desktop app's permission system.
The limitations are real. Your desktop has to stay powered on and connected. Dispatch handles one task at a time. There are no push notifications when tasks complete — you have to check. It is early. It is slow. It is a research preview in every sense of the word.
But the direction is unmistakable. Dispatch is available on the Pro plan at $20/month and the Max plan at $100/month. Anthropic is not treating remote agent control as a premium add-on. They are treating it as a core feature of the platform. That pricing decision tells you everything about where they think this is going.
What It Means for Me
This entire site is built with Claude Code. Every post, every layout, every deployment. I have been running this workflow from my terminal and from the Claude desktop app.
I just connected my phone to my computer through Dispatch. I am not describing a hypothetical workflow — I am living it. The idea that I can text Claude a task from my phone and come back to a finished pull request is not speculative. It is what I did today.
Channels and Dispatch together change the daily rhythm of building with AI. Not the capability — the cadence. The gap between "I had an idea on a walk" and "the code is written" just collapsed to the length of a text message.
The Bigger Picture
Steinberger, the creator of OpenClaw, has since joined OpenAI. Anthropic asked him to rename his project — it was originally called Claudebot. The competitive dynamics here are not subtle.
But OpenClaw is not going away tomorrow. It has a massive community and the open-source advantage. What Channels does is remove the primary reason a Claude user would reach for OpenClaw in the first place: the ability to talk to your agent from anywhere.
Anthropic is saying: you do not need a third-party wrapper. We built it. It is sandboxed. It is secure. It is native.
The moat around OpenClaw just got a lot shallower.
The Protocol: Claude Code Channels lets you message your development agent over Telegram and Discord. Dispatch lets you control Cowork from your phone with a QR code. Both are built on the same thesis: the agent should keep working when you walk away. The real shift is not the messaging — it is the move from synchronous supervision to asynchronous partnership. The bottleneck was never the AI. It was the human's availability to direct it. That constraint just loosened — for developers and non-developers alike.
