Skip to content
SLOT-0704 | 2U RACK

Three Weeks Running My Business With an AI Team

Reading Time
5 min
~200 words/min
Word Count
929
2 pages
Published
Oct 11
2026

Table of Contents

Reading Progress 0%

Since late September, I've been running my business with an AI team. Not a chatbot I ask the occasional question. A team, with roles, a coordinator, and clear rules about what it can and can't do without me.

It's been about three weeks. Here's what got done, roughly what that work would cost if I'd paid people to do it, and what I learned along the way.

How my AI team is set up

At the center is Vision, my orchestrator (a Grok Bot). Vision coordinates nine specialist agents, each with one role:

  • CTO advisor
  • Outbound prospecting
  • Lead desk
  • Inbox tidier
  • Billing watcher
  • Security digest
  • Revenue strategist
  • Software planner
  • Bot designer

For hands-on coding on my workstation, I use Claude Code and Gemini (through the Antigravity CLI).

The agents do the legwork. I still make the decisions.

Sales: 100 businesses, 96 drafts, nothing sent without me

The team researched 100 local small businesses with reachable contacts. For each one, it ran a passive email-security check on the business's domain, looking at three records: SPF, DKIM, and DMARC.

Those records help the rest of the internet tell whether an email claiming to come from your business really did. When they're missing or misconfigured, it's easier for someone to send mail that pretends to be you. "Passive" means reading what each domain already publishes, not poking at anyone's systems.

From that research, the team drafted 96 personalized first-touch emails. Each one names that business's actual finding, not a generic pitch.

None of them went out without my approval.

On the business side, it also built a revenue tracker, a weekly funnel routine, and an offer sheet backed by sourced Midwest pricing research.

Infrastructure: two new services and fewer blind spots

I run a Proxmox homelab, and the team built two new self-hosted services on it:

  • ntfy plus Uptime Kuma for alerts and monitoring. It now watches 13 services.
  • listmonk for an opt-in newsletter.

Backup failures now alert my phone, and the new containers are included in the nightly backups.

The team also found secrets sitting in internal docs and removed them. And it right-sized the instruction files my coding agents read. That second one sounds minor. It turned out to be one of the bigger lessons, which I'll get to below.

Security and admin: practicing what I sell

Before offering an email-security check to anyone else, I audited my own domain. My email has to meet the standard I'm selling.

We also caught a Microsoft 365 billing failure before it took down my business mailbox. For a one-person shop, the mailbox is the front door. Catching that early was exactly the kind of quiet, boring save I want this setup to make.

One more simple habit: I keep a single open-questions doc, so decisions don't get lost in chat threads.

What would this have cost with people?

This isn't the equivalent of a few full-time staff. It's closer to the part-time work of about seven roles: chief of staff, sysadmin, security analyst, sales development rep, copywriter, researcher, and bookkeeper.

To size it, I estimated the hours each piece of work would take a person and priced those hours at local wages. The hours are my own estimates. The wages are Bureau of Labor Statistics mean hourly wages for the South Bend-Mishawaka, IN-MI area (May 2025).

WorkEst. hoursRole (BLS mean hourly)Est. cost
Lead research and domain checks30Market research / marketing specialist ($34.05)$1,022
96 personalized emails15Market research / marketing specialist ($34.05)$511
Two services, monitoring, backups15Network and computer systems administrator ($40.73)$611
Secrets cleanup and doc hardening20Information security analyst ($45.83)$917
Pricing research10Market research / marketing specialist ($34.05)$341
Planning and coordination25Office and admin support ($22.43)$561
Total~115~$4,000

That's about $4,000 in wages. BLS data shows wages are about 70% of total compensation, so the fully loaded cost is closer to $5,700. Using contractors instead would likely run $7,000 to $10,000.

For context, South Bend's mean wage is $27.99 an hour, compared with $33.54 nationally (May 2025). Local labor already costs less than the national average, and the estimate still lands in the thousands.

To be clear, these are estimates. They're a way to size the work, not a precise savings figure.

What running an AI team taught me

Approval gates matter more than autonomy. The most useful thing isn't an agent that acts on its own. Instead, it's an agent that does the work and then stops for my sign-off. Those 96 emails are the example: the team did the research and writing, and I decide what goes out.

Less context makes better agents. It's tempting to hand an AI everything you know. In practice, short instruction files and skills that load only when they're needed work better. I've written more about this in The Context Paradox and Token Economics.

AI usage is a budget. Not every task needs the most capable model. So simple work goes to cheaper models.

Verify everything, and source every number. AI output is a draft until someone checks it. That's why every wage figure in this post has a source, listed below.

What's next

My first milestone is $100k in revenue for my local tech services business.

The plan is the same one that worked these three weeks. The AI does the legwork. I do the relationships and make the final call.

Free email spoofing check for Michiana businesses

If you run a business in Michiana, I'll check your email domain for spoofing risk for free. It's the same passive SPF, DKIM, and DMARC check described above, and I'll explain what I find in plain English.

Request your free email-domain spoofing check

Sources

Eddy Kawira

About Eddy Kawira

Infrastructure Engineer & AI Systems Architect with 20+ years across IT systems, data center operations, and cloud architecture. Experience ranges from high-scale AWS data center deployments (Trainium2) and enterprise Azure Virtual Desktop migrations to building open-source AI agent tooling (ClawHub) used by 10,000+ developers. Runs a battle-tested hybrid Proxmox VE and Hyper-V homelab focused on reliability, deep observability, and safe automation.

View all posts by Eddy Kawira →
user@eddykawira:~/comments$ ./post_comment.sh

# Leave a Reply

# Note: Your email address will not be published. Required fields are marked *

LIVE
CPU:
MEM: