I spent a whole week with OpenClaw.
The hype is real, not gonna lie.
It is brilliant engineering.
But I am not going to sell you a course on it.
What I’m about to say is by no means trashing it.
It IS imperfect, but I don’t fault Peter Steinberger on it.
He has only been building this for 3 months after all.
But I do want to share my OpenClaw experience.
I asked it to do a simple task.
It failed.
Then it asked for permission to "fix" itself.
It wanted to rewrite its own code to solve the error.
I said yes.
It ran the surgery.
And then it went silent.
I messaged it. Nothing.
When I checked the terminal, I found out it was hanging.
It had effectively lobotomized itself.
The process consumed four hours of fighting with dependencies and reinstalling the entire environment three times.
I burned $200 in API credits in a single afternoon.
And even when I finally got it working, I felt a specific kind of anxiety.
I felt like if I typed the wrong command, the whole thing would shatter again.
That is not a business tool. That is a developer's environment.
Right now, the industry is rushing to "wrap" this technology.
They want to sell you the "autonomy."
But to me, autonomy without stability is just expensive chaos.
You cannot build a company on an employee that ghosts you because it tried to do brain surgery on itself.
Which is why I am dissecting OpenClaw instead of wrapping it.
I am using Claude Code to rip apart OpenClaw’s architecture.
We are taking the brilliance - the memory, the planning, the agency.
But we are installing it into NEO instead.
A stable, governed environment.
We are building an engine that runs your business, not a science experiment that burns your credit card.
While I am doing that, I’ve also created a ‘by-product’ that I am already using to help with NEO’s development.
Over the last 48 hours, I built an OpenClaw clone that lives entirely inside Claude Code.
It has the same agency. It does everything the original does. But I stripped out the chaos.
No Hardware Tax: You don't need to buy a dedicated Mac Mini to run it.
No Terminal Warfare: You don't need to fight with code or dependencies.
No Bankruptcy Risk: You run it on your standard $20/month Claude subscription instead of burning thousands in API credits.
It is safe. It is stable. And it is a one-click install.
NEO’s final form will be something better than both.
But right now, this other version is available.
If you join the March Build, you aren't just getting the theory.
You are getting this engine.
I am giving every member access to my personal working copy of this "Safe Mode" agent.
If you want to play, you can go to the repo (it’s free).
If you want to build infrastructure that actually works, join the lab.
We start building on March 1st.
Join The March Build: [Archived link retired.]
FELIX
P.S. I recorded a video breaking down exactly why I made this decision and what it means for the future of our work.
Watch: https://youtu.be/xNKAJ9UlwaI?si=kahIvdFR_jZl7q8X