I was supposed to be writing a LinkedIn post.

At least that was the plan in my increasingly ADD brain—sit down, write the post, move on with my day.

Instead, I clicked a link about OpenClaw. Just to skim it. Ninety minutes, and 3 articles later, I was deep into how someone's AI assistant had called a restaurant, spoken to an actual human being, and booked a table for four — and the AI had set up the phone line itself because it didn't have one yet.

Needless to say, my intended LinkedIn post didn't get written.

You've probably seen OpenClaw in your feed by now. Maybe you bookmarked it. Maybe you did what I do with half the things I bookmark, which is save them certain that I will read them later, and then never look at them again. (I don't even know how many bookmarks I have, too many to count. And I've read maybe eleven.)

But these I actually read. All of it. The docs, the guides, the team write-ups. I haven't set it up yet — I need to be honest about that.

And what I found was enough to sit down and write this instead of my intended LinkedIn post, because I think you need to look at it before you bookmark it and forget.

So What's Actually Going On Here

You use Claude. You use ChatGPT. Personally, I have both open right now in separate tabs alongside 27 other tabs that I can't seem to close. So the question isn't "what is OpenClaw." You know what it is. The question is why it seems different from everything we're already using.

Here are three big things.

  1. It lives in your text messages. Not a browser. Not an app. WhatsApp, Telegram, Discord, SMS — wherever you already talk to people. You text it the way you'd text a friend who happens to be abnormally competent and available at 3 a.m. when you can't sleep. There is no "AI time." There is no switching contexts.
  2. It rewrites itself. This is the part that killed my LinkedIn post. When you ask your Claw to do something it can't do yet — connect to your email, track a flight, check your calendar — it doesn't apologize and suggest you try a different tool. It just writes the code or whatever to make it happen. No plugins. No settings menu. No search where you spend twenty minutes looking for the right connector and then give up and do it manually. You ask. It figures it out.
  3. It works when you're not looking. Most AI tools sit there patiently waiting for you to tell it what to do. A Claw runs in the background. It checks your inbox every thirty minutes. It watches your calendar. It flags what's urgent and stays quiet when nothing is. That last part — the staying quiet — is what separates this from every notification system that ever made me want to throw my phone out the window.

The Thing I Keep Thinking About

What makes OpenClaw "interesting" isn't so much what it does, but what it represents. A personal AI that lives in your messages, runs in the background, and builds its own capabilities when it hits a wall.

To be honest, it feels like we're getting closer to AGI (depending on your definition). Not just another chatbot with extra features. I know I've said this before, it's just a different kind of tool. And OpenClaw isn't the only one doing this.

The Bigger Picture

OpenClaw isn't happening in isolation.

Claude launched Computer Use Agent in late March. It can now open apps on your desktop, navigate browsers, fill in spreadsheets, and run multi-step workflows and you don't need to touch a thing. Message it a task from your phone, it does the work on your computer while you're out living your life. (If that sounds familiar, I wrote about Dispatch in Claude Cowork last month. Same idea.)

KAIROS — this one's a bit crazier. Researchers found a fully built proactive AI agent buried in Claude Code's source code. Designed to run 24/7. Acts on a heartbeat loop. Sends push notifications. Consolidates its own memory while you sleep. Anthropic built it. They just haven't turned it on yet.

Google is building agent extensions into Gemini. Microsoft is doing the same with Copilot.

There's a clear pattern, the next version of AI doesn't need you to open it. It runs in the background, takes action, and checks in when it needs you.

OpenClaw may have gotten there first with an open-source, messaging-native approach. But it's the direction everything is heading. The question isn't whether you'll have a proactive AI agent. It's which one, when, and if you'll be ready to use it well.

What People Are Actually Doing With It

If you're reading this newsletter, the phrase "open-source framework" probably makes you want to close this tab. (I know. I'm not technical either and despite all my reading, testing, etc. — it's a fact that sadly remains true.)

Here's what regular (non-developer) people are doing with their Claws:

Setup requires a terminal command or a server, which is more technical than most of us want for a Tuesday. But there's a hosted version coming — one click, no setup — and that's the one I'm watching for.

The reality is that it's just texting. So basically, we're all already qualified.

My Plan (Sharing So You Can Hold Me To It)

I haven't set mine up yet. Here's what I'm doing later this month, after I leave the Phoenix heat — and I'm telling you so I actually do it instead of bookmarking it with everything else and never looking at it again:

Week 1: Get it running, after I determine what to run it on (the security piece still makes me nervous). Connect it to Telegram. Start simply with a to-do list, a daily check-in.

Week 2: Connect the real stuff. Email, calendar, maybe Notion. Set up a morning briefing. See if "just ask and it figures it out" holds up when you're running four brands, three newsletters, and a downloads folder that should probably be reported to the authorities.

Week 3: Get more ambitious. Scheduled tasks. Weekly planning. And the thing I'm most curious about, whether it can watch my LinkedIn and surface the notifications that actually matter instead of the 47 daily alerts telling me someone I met once in 2019 has a work anniversary (don't you just love the algorithm).

Stay tuned, you know I'll be writing about it (I'm fairly predictable, it's what I do).

Final Thoughts

A few things worth mentioning.

This is real. The proactive model means things happen before you remember to do them. The self-modifying piece means you don't hit a wall and problem-solve every time you need a new integration. And the messaging interface means it actually gets used, because it's where you already are.

That last point matters more than any feature list.

But. Your Claw gets access to whatever you connect it to — email, files, calendar. That's the whole point, and it's also worth being thoughtful about. Start with messaging-only tools. Add more as you go. Same principle as a new hire, give them access to what they need for the job, not the company finances.

It can be confidently wrong. (Sound familiar?) Use it for drafts, research, and organization. Verify anything that matters.

And the security picture deserves a second mention. OpenClaw is open-source, which means it moves fast, so there are real vulnerabilities, including one that let malicious websites hijack a Claw without user interaction. The project patches actively, but this is early-stage infrastructure being used for real tasks. Eyes open.

Third-party add-ons vary in quality too. They have a registry called ClawHub for vetted skills — and the rule is the same as always: if someone in a Discord tells you to install their custom thing, be as skeptical as you would with an unknown file.

The Big Shift

The biggest mistake people make with their Claw is treating it like a search engine. Question in, answer out, move on.

The people getting the most out of it? They treat it like a new hire. They delegate projects. They explain how they like things done. They check the work and give feedback, lots and lots of feedback. Every exchange teaches it more about what you want.

If you've ever managed someone well — given clear direction, useful feedback, built trust over time — you already have the skill set.

This isn't a technology skill. It's a leadership skill.

Which means the people who've spent twenty years managing teams, projects, and crises in complex organizations might be better positioned for this than the twenty-three-year-old who can code but has never delegated anything in their life. No matter what all the articles say, experience matters.

Stay Tuned

I'll be setting mine up this month.

Here's what I think is important to come back to:

We've gone from "AI can answer my questions" to "AI does my work while a ghost drives me across Phoenix" to "AI lives in my text messages, rewrites itself when it hits a wall, and runs my week while I'm at the grocery store."

In less than three years.

And now every major AI company is building toward the same future of proactive agents that work in the background.

So far, it's still not slowing down — and it's exhausting. But the people who figure this stuff out aren't going to have a slight edge, they're going to have a fundamentally different future.

I think that's worth a missed LinkedIn post. Or twelve.

I hope this was helpful,
Andrea