Openclaw Gives Me Anxiety

Revision counter: I asked Claude to fix this post 8 times before it got the details right. Peak AI experience.

I use openclaw daily. I also don’t fully trust it. These two things coexist.

The Anxiety

Openclaw is an open-source AI agent that runs on my machine and messages me on WhatsApp. It’s fun — genuinely useful for quick questions, random ideas, casual back-and-forth. But there’s this low-level anxiety that comes with using it.

It has access to my entire system. It installs packages I didn’t ask for. It occasionally just does dumb stuff with full confidence. And every time it runs a command I didn’t expect, my heart rate goes up a little.

I’m also not entirely sure Anthropic loves that I’m hammering my Max subscription this hard. If someone at Anthropic is reading this — I’m getting my money’s worth. You’re welcome.

Dispatch

This is where things get better. I’ve been using Dispatch — Anthropic’s own agent feature built into Claude Desktop, currently in research preview. You pair your phone to your desktop via a QR code and assign tasks from your couch. It’s noticeably more stable than letting openclaw roam free. The killer feature? You can restrict which apps it has access to.

I hope that actually works. I genuinely do. But even the option existing makes me feel better about letting an AI agent loose on my computer — one that I can talk to from my phone.

The Best Use Case So Far

I ask Dispatch to watch a recipe video short, pull out the recipe and ingredients, and add it all to my Notes app. That’s it. No cooking blog SEO spam, no scrolling past someone’s life story to find out how much garlic to use. Just the recipe, in my notes, ready to go.

That alone justifies the whole thing.

The Vibe

Openclaw in 2026 is like giving your car keys to a teenager. Extremely capable, occasionally terrifying, and you just have to hope the guardrails hold. But when it nails it, you wonder how you ever did things the slow way.