📰 AI News: Anthropic Just Made Claude Code Feel A Lot More “Always On”

🔥

Mar 21 • AI News

📝 TL;DR

Anthropic shipped Channels for Claude Code, letting you control a live session from Telegram or Discord, and Cowork also added Projects on desktop. Together, these updates push Claude much closer to the always on agent workflow people have been craving, where AI keeps working even when you step away from the terminal.

🧠 Overview

This is a bigger shift than it looks. Claude Code is no longer just something you sit in front of and prompt manually, it is starting to behave more like a persistent agent you can reach from the apps you already use.

Channels let outside messages flow straight into a running Claude Code session, while Cowork Projects gives desktop work a more persistent structure. The combined effect is simple, Anthropic is moving from “AI tool” toward “AI operating layer.”

📜 The Announcement

Anthropic has launched Channels for Claude Code in research preview. The feature lets you push messages, alerts, and webhook style events into a live Claude Code session from supported channels, starting with Telegram and Discord.

The key detail is that this is not spinning up a brand new session in the cloud. It sends the event into the session you already have open, so Claude can react in context and continue the same thread of work.

At the same time, Cowork on desktop is becoming more structured with Projects, which helps organize persistent workspaces and makes ongoing agent style workflows feel more practical.

⚙️ How It Works

• Live session control from chat apps - Claude Code can now receive messages through Telegram or Discord, so you can steer an active session without sitting at the terminal.

• Runs through channels - A channel is effectively an MCP server that pushes events into the Claude Code session you already have open.

• Two way communication - Claude can read the incoming event and reply back through the same channel, turning it into a true bridge instead of a one way alert feed.

• Session must stay open - Messages only arrive while the Claude Code session is running, so the “always on” setup works best in a persistent terminal or background process.

• Security by allowlist - Access is locked down through paired sender IDs and allowlists, so random people cannot just start pushing instructions into your session.

• Research preview limitations - Channels are rolling out gradually, currently support only approved plugins, and require Claude Code v2.1.80 or later plus claude.ai login.

• Cowork Projects on desktop - Projects add more structure to desktop workflows, making it easier to keep persistent tasks, files, and goals organized over time.

💡 Why This Matters

• Claude is becoming less terminal bound - The more you can reach your agent from wherever you already are, the more it starts to feel like a real coworker instead of a coding utility.

• This is the “always on” direction - The big story is persistence. AI is becoming something you can leave running, then nudge from chat while it continues working in context.

• Chat apps are turning into control panels - Telegram and Discord are no longer just communication tools here, they become remote controls for live agent sessions.

• Workflow is getting closer to delegation - Instead of reopening a tool and restating context, you can keep one live thread going and steer it as events happen.

• Anthropic is chasing a broader future - This is not only about coding. Combined with Cowork Projects, it points toward an AI setup where workspaces, sessions, and tasks stay alive across time and devices.

🏢 What This Means for Businesses

• Better async workflows - Teams can keep Claude Code running on real work and check in from familiar chat tools instead of being glued to one machine.

• Faster incident response - Alerts, CI results, or other events can be pushed into a running session so Claude can react while engineers are away.

• More realistic agent operations - This moves AI closer to owning long running workflows, where humans supervise and redirect instead of restarting from scratch each time.

• Security needs to stay tight - If you use this, treat it like privileged access, use allowlists, scoped permissions, and only trusted environments.

• Great fit for small teams - Founders, ops people, and lean engineering teams can get more leverage from one live session that keeps working in the background.

🔚 The Bottom Line

Channels for Claude Code and Projects in Cowork may sound like small feature updates, but together they point in one clear direction. Anthropic wants Claude to feel less like a prompt box and more like a persistent agent you can reach, steer, and rely on throughout the day.

That is the real unlock. Not just smarter AI, but AI that stays in motion while you move through your work.

💬 Your Take

If you could control a live AI agent from Telegram or Discord while it kept working on your machine, what would you use it for first, coding help, research tasks, ops monitoring, or something else?

2 comments