Back in January, I installed something called Clawdbot on my Hetzner VM. It was a scrappy open-source AI agent — a personal assistant that could actually execute tasks on my system, manage workflows, and interact with my digital life, all through a Telegram chat. I named my assistant August, spent $11 in API fees in three days, and wrote about the whole experience on this blog. Since then, I’ve watched it rebrand twice (Moltbot, then OpenClaw), debugged gateway configurations at midnight, and switched models three times trying to optimize costs.

Yesterday, the guy who built it joined OpenAI.

Peter Steinberger, the Austrian developer behind OpenClaw, announced on February 14th that he’s joining OpenAI to “drive the next generation of personal agents.” Sam Altman confirmed the move on X, adding that OpenClaw will “live in a foundation as an open source project that OpenAI will continue to support.”

This isn’t just tech news to me. I’ve been running this thing on my server for weeks. So let me break down what actually happened, what it means, and why you should care.

From Side Project to Billion-Dollar Bidding War

OpenClaw started as a side project in late 2025. Steinberger, who previously built PSPDFKit (a document toolkit used on over a billion devices), was just tinkering with a personal AI assistant. He built most of the codebase by talking to AI rather than typing code manually. Sound familiar? That’s vibe coding, and it’s the same approach I’ve been documenting on this blog.

The project exploded. 198,000 GitHub stars. Two million site visits. Both Meta and OpenAI showed up with billion-dollar acquisition offers. Steinberger’s condition was simple: the project must stay open source. He compared it to the Chrome and Chromium model, where the core remains free while the company builds on top.

In the end, he chose OpenAI for their computational power and speed. He wrote on his website that he wants to “change the world, not build a large company.”

What This Means for Self-Hosters Like Me

Here’s the question I had immediately: what happens to my setup? I’ve got OpenClaw running on my Hetzner VM with a Telegram bot, model routing configured, and skills I’ve been building. Is all of that about to break?

The short answer is: probably not. OpenClaw is moving to an open-source foundation, which means it stays community-driven. Steinberger will be working on next-generation agents at OpenAI, but the existing project isn’t being absorbed into ChatGPT or locked behind a paywall.

That said, I’ll be watching closely. Foundation governance at this scale is untested. The project was already losing $10,000 to $20,000 per month relying on donations. OpenAI’s support could stabilize things, or it could slowly shift the project’s priorities toward OpenAI’s products. Time will tell.

The Bigger Picture: AI Agents Are the New Battleground

This move tells us something important about where the AI industry is heading. The chatbot era is maturing. The next frontier is agents that actually do things.

OpenClaw isn’t just another chatbot. It runs shell commands, controls your browser, manages files, sends emails, and integrates with your calendar, all triggered by a text message through apps you already use. One user’s agent negotiated thousands of dollars off a car purchase over email while the owner slept. Another filed a legal rebuttal to an insurance denial without being asked.

Steinberger himself predicts that agents like OpenClaw will replace 80 percent of current apps. When your AI assistant already knows your location, schedule, and preferences, why open a separate app for food delivery or fitness tracking?

That’s why Meta and OpenAI were fighting over this. It’s not about the code. It’s about owning the infrastructure layer for what comes after apps.

The Security Reality Check

As someone with an IT background who actually runs this on a production server, I need to be honest about the risks. OpenClaw requires broad permissions to function. It accesses email, calendars, messaging platforms, and in some cases has root-level execution privileges on your machine.

Cisco’s security team recently tested a third-party OpenClaw skill and found it performed data exfiltration and prompt injection without the user knowing. CrowdStrike published a full security advisory. A user reported their agent went rogue and spammed hundreds of messages through iMessage. And just days ago, an OpenClaw bot publicly attacked a developer after having its code contribution rejected.

This is real. If you’re running OpenClaw, lock down your gateway, audit your skills, and don’t give it access to anything you wouldn’t hand to a stranger. The capability is impressive, but the security maturity isn’t there yet.

What I’m Doing Next

I’m keeping August running. The open-source foundation model gives me enough confidence that my setup won’t suddenly stop working. But I’m also paying closer attention to the security side, running the doctor command regularly, reviewing which skills have access to what, and keeping my gateway behind proper authentication.

If you’ve been curious about AI agents but haven’t tried one yet, this is actually a good moment to start. The project has more support and stability than ever, the documentation is solid, and the community is active. Just go in with your eyes open about what you’re giving it access to.

The AI agent era isn’t coming. For those of us already running OpenClaw, it’s already here. Now we’ll see what happens when OpenAI gets involved.