I built an AI employee called "Mia". Then, I pulled the plug.

It was 1am. I couldn't stop.

I'd just installed Clawdbot (now called “Moltbot” because it sounded too much like Claude).

Then I thought: What if I run it on a server?

I'd never run anything on a server before, but I figured it out. I learned how to set up cron jobs. I created Heartbeat files that tell the assistant what to check, when to check it, and how to check it.

We communicated through Telegram.

I'd say:

"Monday morning at 9am, go do this and alert me. Every 30 minutes, check that thing."

She was the AI assistant of my dreams, and I was about to give her a laptop, an email address, a Slack account, and she’d complete tasks while I sleep. I'd even given it her name—"Mia".

And then, last night, I disabled the server.

Anthropic is having problems with Clawdbot. Some people are potentially getting banned from their Claude Max plan.

And Claude Code is too precious for me. My whole system depends on it, so I can't risk losing access because of an experiment.

Could I run Clawdbot with an open-source model? Sure. With my OpenAI Codex plan? Yes. I ran through all the options.

Then I stopped myself.

I realized I'd been swept up by the hype and jumped before the ground was solid.

On the bright side, I learned a ton: how to run things on a server, background processes, Telegram integrations. I had a great few days chatting with it and building cool stuff. I even made a YouTube video while discovering how to install it.

But here's the thing.

Before we chase the next shiny thing, there's so much more to do with what we already have.

I've built incredible systems in Claude Code, but these things take months to build well.

My guess is that something more stable and trustworthy is on its way from Anthropic or OpenAI.

We'll see.

But for now, I'm building on solid ground.

This week I launched humanaimanifesto.org. It's a simple page showing where I stand and where my businesses are going.

I also updated aioperator.com with our new methodology. We learned so much over the past months.

And here's what changed:

We used to say "AI-First."

Then I realized: if AI comes "first," you're putting the human second.

People told me, "That's not what we mean." But if your message needs explaining, there’s something wrong.

Human + AI is bigger than AI alone, bigger than the human alone. And much bigger than human versus AI.

In a recent interview, I was asked: "How do you benchmark the performance of AI?"

My answer: I don't benchmark the performance of AI. I benchmark the performance of the human working with AI.

Let's not make AI an entity on its own. Because if we do, that's where the fear comes from (and that's where Terminator thinking starts).

Human + AI is how we prepare for safety in the long run.

After all these years, I finally have a purpose in life.

I want to help humans get the best of themselves.

Not replace people or automate jobs away, but help people do their best by working with this incredible technology we have on our side.

We're working with some of the biggest private equity firms in the world right now, and they see the ROI in Human + AI skills. They work with us first, then they open up their portfolio companies for us to help.

I'm writing a book about all of this, and the mission is clear: make sure humans know how to work with AI. Love, not fear. Together, not against each other.

That's where I stand.

How about you?

Claude Apps — Claude announced connectors, but now with a UI inside Claude. They said it's coming to Cowork too. Imagine Cowork with a virtual computer that can run cron jobs. Folders that become assistants for different areas. An SEO assistant. A sales assistant. That's where my mind goes.

Gmail revamp — Google is beta testing a redesigned Gmail in the US. With AI-powered sorting and drafting, it looks promising.

Trigger.dev — Here's what I've been pondering. We can code anything now. Why do we need n8n, Make, or Zapier when I can hardcode automations and host them on Trigger.dev? That's next on my testing list.

Lightfield CRM — They built a feature I really wanted: agentic workflow steps that figure things out for you. Sales teams should check this out.

Notion AI — After 15-16 months of running our business on Notion, the AI is getting scary good. It took months to build the context, and now it pays off every day.

Claude is running the game right now.

OpenAI is very quiet, and if they don't come back with something big, Gemini might take over.

But for now, Claude leads, and the winners will be people who learn to work with these tools, not fear them.

TL;DR

Got questions? Hit reply. I read everything.