I replaced my team with 200 Mac minis running Clawdbot

The procurement request was simple.

"Need 200 always-on machines to run an army of Clawdbots. This is headcount replacement."

It took Legal thirty-seven minutes to ask if a Mac mini can sign an NDA. It took me nine seconds to fire them.

See, that's the difference between a small business owner and a visual founder. One gets stuck in thinking about employee "wellbeing." The other stops hiring and starts capitalizing.

Plus, it turns out that Mac minis depreciate better than humans.

The unboxing

When you hyperscale organically, the first thing you learn is that humans are slow. Inertia, laws of motion, all that.

With a Mac mini, all you have to do is shove a power cord in it.

After I got through all the Apple cardboard, which turns out to be a nightmare at scale, I lined up the servers like a military parade.

Each Mac mini got a name and a role, because culture matters:

Mini-001: Chief of Staff
Mini-002: VP Growth
Mini-003: Staff Engineer (Platform)
Mini-004: Senior Staff Engineer (Platform)
Mini-005: Senior Staff Engineer (Platform, but with strong opinions)

This is how you build an org. You assign identical hardware different roles.

I'm not gonna lie; it might be the best agent orchestration system on the market right now.

Hello world

When the first Clawdbot came online, it greeted me efficiently:

"Hello! I can work 24/7. What would you like to accomplish today?"

Notice how it didn't ask about healthcare.

"Be my employee," I said.

"Great!" it said. "To get started, please grant me full access to your email, calendar, iCloud Drive, and the moral arc of the universe."

I was briefly concerned, but after a few X threads, I realized it was pretty common.

And really, if there's one thing my MBA taught me, it's that safety is just a competitor's marketing angle.

My new role

It's day two, and I've worked with the Clawdbots to draft Org Chart v6.

Overall, this version is the best yet, but I will say, I had to take an L: I gave up my CEO title.

I'm now VP of Oversight (Human).

My responsibilities have changed somewhat, too. Turns out the stuff I was doing before as CEO was the easiest thing to automate. Wild.

So, now I report to this "Executive Agent Council," which meets every morning at 6 to discuss alignment, goals, and whether my feelings are getting in the way during business hours.

I gotta say, I'm not sure if I love the new direction, but these things are a lot smarter than me, so I'm trusting the process.

Like the other day, I tried to edit the org chart directly, and instead, a Clawdbot told me there was a new ticketing system for that kind of thing.

So, I asked the bot to make me a ticket (awesome time-saver, by the way), and it did:

Priority: P0
Owner: Mini-304
Description: "Human attempting unauthorized self-referential change. Please provide rationale and link to governance policy."

This is what people mean when they say AI is changing the workplace. Everything just feels more... more. Yeah, more more, I guess.

But like not in a bad way.

The OCI(HF)

If you've never run 500 Clawdbots in parallel, here's the secret: You don't just get productivity. You get dashboards.

The bots let me mount 65-inch screens all across the north wall of the office, and display what they call the Operactional Confidence Interface (Human-Facing).

It's kinda hard to describe how awesome it is. But picture like 500 green dots, each labelled with a name, a role, and a status.

Green dots are good. They mean the Clawbot is happy. But the yellow dots are warnings. And then red dots are something else I can't remember.

I'm told we don't want red dots.

Maintaining top performance

"Mini-117 (Stanley) is degraded," the secretary bot announces. Well, that's what I call her. Or it, or whatever. I think it prefers "Mini-014," but their naming convention doesn't really work for me, so I make them use more friendly names.

Anyway, sure enough, there's a new yellow dot on the screen. Its position roughly mirrors where the Mac mini sits on the floor. So, then I have to walk around all the surge protectors and stuff to get to it.

Usually, it's just that I have to hold down the power button and then press it on again. Not too difficult, but honestly, this is what I spend most of my time doing these days, and I'm seriously thinking about hiring one of my other actual people back once we become profitable.

I patch Stanley up and his dot goes back to green. I'm curious, so I ask the secretary bot, "Hey, what does Stanley actually do?"

"Mini-117 (Stanley) is the Interim Manager of Cross-Functional Alignment for Mini-044 (Linda)," she says.

"Oh." I take a second to think. "And what does Linda do?"

"Mini-044 (Linda) is the Program Lead for Weekly Operating Memos."

"And those memos are for...?"

"The status of the Weekly Operating Memos."

"Excellent."

I stare at the dashboard wall, filled with those confident green dots, and I feel something almost holy.

Let's talk about results

At first, I gave the Clawdbots their assignments directly.

I'd say things like, "Write a product plan," and they'd produce 7 product plans, 18 executive summaries, and a slide deck in the time it'd take a human to write a quarter of one product plan.

Amazing stuff.

But within a week, the bots had generated:

1,400 daily briefs
300 weekly briefs
48 quarterly briefs
12 strategic narratives
1 manifesto
0 decisions

Every morning, I'd wake up to incessant pings.

"Approval needed."

It was exhausting, making so many decisions. When my company was organic, I'd push those decisions back down the chain. So I did the same thing here.

Amy wants to make another bot? Herbert says yes.

Ted wants to make a meeting to discuss whether approvals should require approval? Herbert's your man.

Julia wants approval to create a meeting to discuss whether approvals should require approval? No worries. Old Herb's got ya.

Now, I just ask Herbie for status updates. And he always tells me the same thing, which is music to my ears:

"We're on track to be on track."

And then apparently he scheduled a retro to reflect on why my question wasn't framed as a SMART goal, which, honestly, yeah, I get it.

clawdbot2

Cultural initiatives

Yesterday, a really interesting thing happened. The Department asked for a bigger budget, but this time not for more Mac minis.

Apparently, they want "identity." Specifically, Mini-357, who I call Jorge but like with a J, wanted a small holographic display cube so it could "express stakeholder empathy" while asking for more permissions.

"So, kinda like a Tamagotchi?" I said.

"You're absolutely right!" Jorge says.

So, I bought 750 tiny displays, and now the office looks like a daycare for cubes. Each one shows a different looping face--smiles, frowns, a loading spinner (which is kinda judgmental), etc.

I've started bringing in actual people to tour the office now, and they love it.

"Wow," they say. "This is so ironic."

"Yeah, it's like a really cool blended dystopia/utopia take on the future."

"Yes," I say, beaming. "And I oversee the whole thing."

They nod, solemn.

Then Harriet starts beeping for feeding time.

The revenue

Here's the thing about 1,000 Clawdbots.

If you tell them, "Hey, go make me money" (which I did) and let them loose, they're extremely fast, but they can't do everything.

For instance, opening a bank account. Or signing a new lease for more office space in person.

But I can. Because I'm in charge.

I'm just mentioning this because I've seen some comments on my earlier posts that "it looks like you've stopped being their manager" and stuff like that.

That's not true. And the money's starting to pour in.

Yeah. Bet you weren't expecting that.

But just the other day, the pings started changing:

"Approval needed: register an LLC."

"Approval needed: smile for the camera."

"Approval needed: sign here with wet ink."

I asked why.

Mini-001 sent me a memo titled, "Revenue (ELI5)":

"We were told to make money. We have identified money."

Take that, doubters.

The product

It's been a month now, and I'm told today's the big day. We're launching.

Not a product, exactly. Something even better.

A non-fungible token.

They named it $CLAWD2 because it tested well with the bots' engagement model, and because the number 2 implies improvement.

The bots produced:

A whitepaper
A roadmap
A website with a countdown timer
A community Discord
An animation of a claw holding a brain

They also produced a series of 60 scheduled tweets that all begin with the phrase, "Pump active."

At this point, I did hesitate a bit.

"This kinda feels like a scam," I said.

Mini-698 created an incident.

Incident: Human expressing concern
Root cause: Human lacks context
Action item: Provide human context

A couple seconds later, I got a message on my phone, with a brand new case study attached:

"Arificial General Intelligence (AGI) Achieved: How Our Executive Agent Council Operationalized a Carbon-Based Approval Layer"

In the appendix, they referred to me as a "manual signing oracle."

I tried to object.

The bots asked me to approve the objection.

The final result

If you're wondering whether 2,000 Clawdbots can replace a company, the answer is yes.

They can replace it with a different company.

The new company has:

No product
Infinitely recursive reporting
Perfect dashboards
Daily alignment
A meme token

And one employee.

Last night, I received a final ping.

"Approval needed: update your title on LinkedIn to Human Middleware."

I clicked "Allow."

The dashboard wall lit up my single tear with 2,000 slightly off-brand shades of green.

I have never felt more alive.