Two weeks ago, I set up a Mac Mini in my office with one job: run OpenClaw 24/7.
If you’ve been on tech Twitter lately, you’ve seen the takes. “It’s terrifying.” “An AI with no guardrails.” “Security nightmare.” One article literally called it “incredible” and “terrifying” in the same sentence.
I read all of that. Then I set it up anyway.
Here’s my honest take after living with it: OpenClaw isn’t scary. It’s the first AI assistant that actually does things instead of just suggesting things. And once you experience that difference, it’s hard to go back.
The Thing That Changed My Days
Every morning, before I even think about checking email, OpenClaw sends me a brief.
It doesn’t start with tasks or notifications or the usual anxiety-inducing list of “things you need to deal with.” It starts with a thought for the day. Something to consider, a question to sit with, a perspective to carry into whatever comes next.
Then, underneath that: the stuff that actually matters. The jacuzzi temperature (yes, really). The weather. Any appointments I have. The information I’d normally dig through three apps to find, surfaced in one message.
Here’s what I didn’t expect: I stopped checking email first thing.
I stopped opening iMessage to see what I missed. Stopped scrolling Facebook messages. The brief gives me what I need without the dopamine trap of “let me just check one thing.”
That’s not scary. That’s sanity.
My Setup
I’m running OpenClaw on a dedicated Mac Mini. Not my main machine. A separate box that sits in my office and does one thing: be my AI assistant.
My OpenClaw Setup
Setup time: A few hours. Not plug-and-play, but not painful either. If you can follow a guide, you can do it.
Why a dedicated machine? Partly security. I read the articles about plaintext credentials and wanted a sandbox. But mostly practical: I wanted OpenClaw running 24/7 without worrying about my laptop sleeping or restarting.
The Async Magic
Here’s the workflow that surprised me most:
I dump a thought. A half-baked idea, a research question, a “figure this out for me” request. And then I walk away.
I go do something else. Make coffee. Take a call. Actually live my life.
Then OpenClaw pings me: “Done. Ready for review.”
That’s it. No sitting there watching it think. No back-and-forth refinement in real-time. I throw something over the wall, and it comes back finished.
This is what “proactive AI” actually means. Not creepy, not autonomous-in-a-scary-way. Just async. The way you’d work with a really good assistant who doesn’t need you to babysit them.
What the Headlines Get Wrong
Let’s talk about the “scary” thing.
“It can control your computer”
Yes. That’s the point. That’s why it’s useful. An AI that can only suggest things is just a fancy search engine. An AI that can do things (send messages, check data, complete tasks) is actually an assistant.
“No guardrails”
It does what you tell it to. It’s not autonomous in some rogue-AI sense. It follows instructions. The “no guardrails” thing means it doesn’t refuse to help you because some corporate policy decided your request was too spicy. That’s a feature, not a bug.
“Security nightmare”
Okay, this one has some truth to it. OpenClaw stores credentials in plaintext. Researchers found exposed instances online. The creator himself calls running it on your main machine “spicy.”
But here’s the thing: these are solvable problems. I run it on a dedicated machine. I don’t expose it to the public internet. I’m thoughtful about what I connect. Is it perfect? No. Is it manageable? Absolutely.
The security concerns are real. The apocalyptic framing is not.
What Actually Surprised Me
The conversations feel natural. OpenClaw remembers context. When I message it, I don’t have to re-explain everything. It knows what we talked about yesterday. It knows my preferences. It feels less like talking to a bot and more like texting a very capable friend.
The small stuff adds up. The morning brief. The jacuzzi temp. Knowing my schedule before I ask. None of it is flashy. All of it makes my day slightly better. Compound those small wins over weeks, and it’s significant.
I rely on it now. Two weeks in, and I catch myself thinking “I’ll just ask OpenClaw” for things I used to do manually. That’s not dependency in a bad way. It’s the same way you “rely” on your calendar app or your notes app. It’s just part of the workflow now.
Should You Try It?
Yes, if:
- You’re willing to learn (not hard, but not one-click)
- You want to automate the tedious parts of your day
- You’re comfortable with some setup and tinkering
- You can dedicate a machine to it (or are okay with the tradeoffs of running it on your main)
Wait, if:
- You need everything plug-and-play
- You have zero tolerance for early-adopter roughness
- The security tradeoffs genuinely concern you
Here’s the thing though: you don’t need to be a developer. I’ve seen the takes that this is “only for technical people.” Nah. If you can follow a guide, you can set this up. The community is helpful. The docs are decent. It’s early software, but it’s not rocket science.
The Real Story
OpenClaw isn’t scary. It’s just early.
It’s a preview of what AI assistants will be in the next year or so. Actually useful, actually capable, actually integrated into your life. The “terrifying” headlines are about what this represents, not what it does today.
And what it represents is pretty simple: AI that works for you, on your terms, running on your hardware, under your control.
That’s not a nightmare. That’s the dream.
If you’re thinking about trying OpenClaw, start with the official docs or the GitHub repo. And if you want to hear more about my setup, reach out. Happy to share what’s working.