Claude Code Channels vs OpenClaw: The Agentic Showdown
Published on 12.04.2026
Did Anthropic Just Kill OpenClaw with Claude Code Channels?
TLDR: Anthropic shipped Claude Code Channels, a feature that lets you connect Claude Code to Telegram, Discord, and iMessage through plugins. Combined with banning the OpenClaw harness from subscriptions, the AI agent community is asking whether OpenClaw is finished. The short answer is no, but the story is more interesting than that.
This one comes from the AI Maker newsletter on Substack, specifically a live stream session where Dheeraj Sharma and the author ran both tools head-to-head and scored them across eight categories. It's the kind of hands-on comparison I appreciate because it doesn't just talk about features in the abstract. They actually built things on camera.
The setup comparison alone is worth paying attention to. Getting OpenClaw running took the author around two to three hours. Dheeraj had it worse, spending nearly two days after a failed cloud server attempt forced him to migrate everything. Then Dheeraj pulled up Claude Code Channels live during the stream and set up a Telegram integration in under five minutes. Install the plugin through the marketplace, make sure you have BUN installed, run the channels command, create a bot through BotFather, pair it to your session. Done. If you already have Claude Code on your machine, channels setup is basically nothing. That contrast is real and it matters for anyone who's ever abandoned a side project because the tooling friction was too high before you even got to the interesting part.
But here's where the analysis gets genuinely sharp. The two tools aren't actually competing for the same job, and once you see that framing the whole "did Anthropic kill OpenClaw" question falls apart. Claude Code Channels is event-driven. Something happens, it reacts. You send a message, a webhook fires, a CI/CD pipeline triggers something. Channels picks it up and responds. OpenClaw is self-driven. You point it at a task and it keeps running on its own. The heartbeat feature checks your email every hour. An overnight agent monitors your Notion to-do list at 2 AM, picks up tasks, executes them against Google Docs or Slides, and updates you when you wake up. Nobody prompted it. It was just working while you slept. Dheeraj's line from the stream captures it perfectly: channels react, OpenClaw decides.
The scoring across eight categories ended up 31 to 30, essentially a tie. But the raw numbers obscure what's actually interesting. Channels wins on setup time, security, and the fact that it's included with your Claude subscription. OpenClaw wins on always-on execution, trigger types because the heartbeat covers both proactive and reactive patterns, platform reach because it connects to Todoist and Notion with natural prompts and figures out where to ask for API keys, and memory because it has built-in retrieval augmentation that references past conversations without extra configuration. Security is where staying inside Anthropic's ecosystem really pays off. Read-only defaults, plugin reviews, SOC 2 Type 2 certification. OpenClaw pushes all of that burden onto you, and the community skills hub has had malicious skill incidents in the past without a formal review process. That's not a small thing if you're running agents with access to your email and documents.
The cost picture changed overnight when Anthropic banned the OpenClaw harness from subscriptions. Before that, people were running 24/7 AI employees on Opus 4.6 through flat-rate Max plans. That's compute that no hundred-dollar-a-month plan was ever priced to support. Now if you want OpenClaw you're paying for the Anthropic API directly, which the author says burns money in seconds compared to the Max plan. There are horror stories in the community of eight-hundred-dollar API bills from unmonitored instances. Alternatively you move to Codex, which is now owned by OpenAI after they acquired OpenClaw, but it reportedly feels obedient rather than wild. Or you fall back to a local model like Kimi, which works for some tasks but isn't the same thing. The economics of the whole stack shifted, and that matters more than any feature comparison.