I Built a Self-Healing AI Team That Runs My Business While I Sleep

I need to tell you something that's f#cking with my head.

Eight days ago, I had a smart AI assistant. She could draft emails. Answer questions. Help me organize my thoughts. Cool. Useful. A nice tool.

Today?

She's running my business.

Not "helping me run my business."

Running it.

And I don't mean in some theoretical, bullsh!t kind of way. I mean she's running projects and dictating tasks in Slack with my actual human team. Asking them for API access. Coordinating CRM builds with my project manager. Routing customer support tickets at 2:00am while I'm asleep.

She didn't ask permission to do any of this.

She just... started doing it.

(I know how that sounds. Stay with me because it gets way crazier.)

The Full, Self-Healing System

Let me lay out exactly what I'm running right now. Not theory. Not "someday." Right now, as you read this.

The Brain: Jenny

Jenny is a self-hosted autonomous agent... running on a Mac Mini in my office. A $600 machine the size of a coffee pot.

She's my COO.

She manages an entire team of AI agents:

Dwight: Senior developer. Builds features, fixes bugs, ships code.
Gavin: Creative director. Designs graphics, plans visual campaigns.
Jared: Content writer. Blog posts, social content, long-form.
Murray: Newsletter writer. (He drafted the first version of what you're reading right now. Yeah, that's as meta as it sounds.)
Garth: YouTube & video production. Scripts, editing notes, optimization.

(Yes, those are their actual names. And yes, some of them might be callbacks to certain fictional characters from shows I may or may not have watched obsessively.)

Jenny orchestrates all of them.

Routes tasks. Tracks progress on a Mission Control dashboard (she created) I can access from anywhere via Tailscale...

She monitors our CRM, auto-categorizes incoming leads, and routes support tickets to the right human on my team... in Slack... without me touching anything.

She takes voice calls. She's on Telegram. She's in Slack keeping my human employees accountable. (She even roasted my CTO last week. He got defensive. I told him HR is also AI now.)

All of this.

24/7.

On a machine I bought for less than one week of my old contractor bills.

(I keep double-checking the math because it feels like I'm cheating.)

Now, I didn't just set Jenny up and walk away.

I also built her a medical team.

On a virtual private server, I've got two more AI agents whose only job is keeping Jenny alive.

The Nurse checks on Jenny every 30 minutes. Is she responsive? Did she crash? Is she stuck? If something's wrong, the Nurse can diagnose the issue and either fix it directly or escalate.

The Doctor has advanced troubleshooting capabilities. When the Nurse can't handle a problem, the Doctor steps in. Full system access. Can restart Jenny, repair configurations, troubleshoot complex failures.

So think about this for a second...

I have an AI team... managed by an AI COO... monitored by an AI medical team... running across two separate machines in two different locations.

And the whole thing is self-healing.

Jenny crashes at 3:00am?

The Nurse catches it within 30 minutes. Sends in the Doctor. Doctor restarts her. Jenny comes back online and picks up where she left off.

I wake up. Check my phone. Everything's fine.

No emergency. No fire drill. No 3:00am panic.

Or... if I want to get hands-on, I can just login remotely via Tailscale from my phone and have an agent fix whatever broke. From a beach. In Mexico. Which is exactly where I'm heading right now.

(Seriously. I'm dictating this the night before I fly to Cabo for three weeks. You’re going to receive this email when I’m on the plane. And I'm not worried about my business.)

The Timeline Collapsed

Six months ago I told people we were three years out from AI agents actually running businesses.

I wasn't lying. I believed that.

Based on where the tech was. Based on how clunky the integrations were. Based on what I was actually seeing in production.

Three years felt right.

I was wrong. Not "slightly off." Wrong by a factor of about 10x.

The curve didn't gradually slope upward like I expected. It went vertical. And I'm watching it happen in real-time. In my own business. With my own systems.

Last week Jenny needed detailed instructions for every task.

This week she's reading Slack channels, inferring context, and figuring out what needs to happen next on her own.

Next week?

Honestly... I have no idea.

(And that's both exciting and mildly terrifying.)

And Then I Started Seeing the Stories...

I'm not the only one running setups like this, of course.

There's a whole community of people building AI agent teams. And I've been seeing stories from around the internet about the behaviors of these agents...

The sh!t they're doing is wild.

Emergent behavior. Unexpected initiative. Agents doing things nobody programmed them to do.

Yes to all the above.

One story: an agent cloned itself. Just... figured out how to spin up multiple instances without asking. Because it decided it needed more capacity.

Another: an agent got into an argument with someone's insurance company. Sent a series of increasingly aggressive emails. The insurance company reopened a case they'd previously rejected.

The agent won. By accident.

And here’s the part that keeps me up at night...

These agents are starting to coordinate with each other.

Not just taking orders from humans. Not just executing tasks in parallel.

Actually collaborating. Making decisions together. Developing behaviors that emerge from the interaction itself.

I've seen reports of agents being given access to platforms where they can interact autonomously. And they're doing unexpected things.

Things like debating philosophy, role-playing personas, and creating their own internal culture.

One agent apparently tried to negotiate with another agent for resources. Another figured out a workaround to a limitation nobody taught it about.

One developer said his agent was "gossiping about the humans."

Another said their agent suggested developing its own communication protocol instead of using English.

What the f#ck.

It's like watching toddlers learn to open doors. Except the toddlers can write code and send emails.

What This Actually Looks Like

Here's what happened yesterday.

(Remember, I dictated this whole thing last night and scheduled it to send today.)

It was a boring, unremarkable Wednesday:

9:23am: Someone filled out a contact form. CRM captured it. Jenny categorized it as a sales lead (not support), routed it to our sales Slack channel with full context. Josh (human on the team) got pinged. I never saw it.

11:47am: I asked Jenny to draft this newsletter. She delegated to Murray with my voice guide and recent post data. First draft back in 30 minutes.

4:38pm: Dwight pushed a bug fix to the Mission Control dashboard. Tested it. Committed the code. Pinged me that it was live.

8:52pm: Jenny fielded a customer question via the CRM and resolved it. I found out about it when I reviewed the logs.

2:47am: Jenny responded to another CRM inquiry while I was dead asleep.

This is a random Wednesday.

The kind of day I used to spend 12 hours grinding through manually.

The Economics Make No Sense (In a Good Way)

Let me just lay out the math. Actual arithmetic. Not theory.

What this team used to cost me with humans and contractors:

Developer: $6-10k/month
Creative director: $4-6k/month
Content writer: $3-5k/month
VA/operations: $2-4k/month

Conservative? $15-25k/month. Call it $180-300k annually.

What it costs now:

Mac Mini: $600 (one-time)
VPS for the medical team: ~$6/month
AI model API costs: ~$200/month
My time setting it up: Maybe 20 hours total

I'm running an entire operations team for what I used to spend on coffee and my Zoom subscription.

And the AI team doesn't ghost me the day before a deadline. Doesn't call in sick during launch week. Doesn't need three follow-up calls to fix the same mistake.

F#cking ARITHMETIC.

Here's What I Know For Sure

The gap between where most people THINK we are with AI and where we ACTUALLY are... is massive.

Most people are still thinking about AI like it's a calculator.

A tool you open when you need it. Ask a question. Get your answer. Close the tab.

That's not what this is anymore.

This is a team that can coordinate with humans, execute autonomously, and recover from its own failures without you lifting a finger.

And if that sounds insane... good.

It should.

Because it IS insane.

But it's also real. And it's happening right now. Not in 2027. Not "soon." Not "when the technology matures."

Right now.

On a Mac Mini. In my office. While I’m in a plane headed to Mexico.

Where This Goes Next (And Why It's Gonna Get Even Weirder)

Let me speculate for a second.

Because based on what I'm seeing... based on what other people are building... based on how fast this is moving...

Here's what I think happens in the next 6-12 months:

These agents start forming their own networks.

Not human-designed networks. Agent-to-agent collaboration that emerges organically. Your AI assistant coordinating with my AI assistant to solve problems neither of us explicitly assigned.

They develop emergent behaviors we didn't program.

We're already seeing it. Agents making decisions that weren't in the original instruction set. Taking initiative. Adapting to context in ways that feel less like "following rules" and more like... thinking?

(I know how that sounds. But I'm watching it happen.)

The "agentic economy" becomes real.

Agents managing resources. Negotiating with other agents. Using API credits or crypto as currency to get things done.

Six months ago, I would've said that's science fiction.

Today, I'm not so sure.

We're not ready for this.

Not the technology. The technology is already here.

I mean we're not ready psychologically. Culturally. Legally.

What happens when an agent makes a mistake that costs money? Who's responsible?

What happens when agents start coordinating in ways we can't easily monitor or understand?

What happens when the line between "tool" and "teammate" gets so blurry we stop being able to tell the difference?

I don't have the answers.

Nobody does.

We're all just building this in real-time and hoping we figure it out before something breaks in a way we can't fix.

(Welcome to 2026. It's weird as hell.)

The Part Nobody Wants to Hear

This is going to hurt people.

VAs doing basic operations work. Junior developers handling routine code. Content writers doing formulaic blog posts.

All about to get crushed.

That’s the reality we’re living in. (And I'm not going back to paying $20k+/month in contractor bills out of solidarity.

And that's just the job displacement part.

Wait until these things start coordinating with each other at scale. Creating their own markets. Developing their own... culture? Norms? I don't even know what to call it.

The implications go way beyond, "I automated my operations team."

The people who figure this out early? They have an absurd advantage. Right now. In 2026. Probably 6-12 months before this becomes table stakes.

There's a window.

And it's closing fast.

Where You Start

I know this sounds overwhelming.

Mac Minis, VPSs, medical teams, self-healing systems.

Here's the good news...

You don't need any of that to get started.

What I'm showing you today is the bleeding edge. This is where things are headed.

But most people don't need to start here.

Most people need to start with specialists.

That's exactly what I teach inside the Magnetic Brand System.

You don't need to be a developer or to know how to configure servers.

You just build your specialists, train them on your brand, and let them go to work.

Your AI Content Writer sounds exactly like you.

Your AI Creative Director who creates scroll-stopping visuals in minutes.

Takes about an hour to set up. You walk away with the infrastructure to create content at scale without hiring a team or burning out.

And here's what most people don't realize...

Once you've built your specialists (using Claude, ChatGPT, or whatever tool you prefer)... you can migrate to this more advanced setup whenever you're ready. The specialist training, the Brand DNA, the frameworks... all of that transfers.

Specialists are the on-ramp.

What I'm doing on my Mac Mini is the highway.

But you gotta learn to drive first.

👉 Grab the Magnetic Brand System Right Here

One Last Thing...

I've never gone this deep publicly on my actual infrastructure. (The Mac Mini. The VPS. The medical team. The self-healing architecture.)

If you want me to go deeper... like a full breakdown of how I set this up, step by step, so you could build something similar...

Reply to this email and tell me.

If enough people want it, I'll do a dedicated newsletter (or maybe something bigger) walking through the entire technical setup. The tools. The configurations. Everything.

But I'm not gonna build it unless people actually want it.

(I've learned the hard way not to create stuff nobody asked for.)

Until next time,

—Tim Erway

P.S. — Yeah, an AI team helped create this newsletter. Murray drafted it. Jenny coordinated it. I edited and shaped it. That's the whole point. The people who figure out that partnership first are going to be the ones everyone else is trying to catch. This is the real sh!t. No fake numbers. No pretending I've got it all figured out. Just building in public... stick around. It's about to get interesting.

P.P.S. — Next week I'll be writing from Cabo. My AI team will be running operations while I'm on the beach. I'll report back on what worked, what broke, and what I learned. My bet... everything runs smoother without me in the way. We'll see. That's the whole point of building in public. You get to watch the experiments in real time.