My AI Journey Was Never About AI. It Was About Marketing Systems.
This blog reads like it's about building AI agents. It isn't, not really. Every crash story and guardrail here was a piece of a marketing system I build for real estate and lending operators.
If you’ve read the posts here, you know the shape of them. An orchestrator that wrote 549 GB and panicked my kernel. A fence that assumes the agent is hostile the moment it reads its first email. A pixel I swore was firing, that wasn’t.
It reads like a blog about building AI agents. It isn’t, not really.
Every one of those was a piece of a marketing system I was building for real estate and lending operators. The AI was the tool. The system was the point. I’ve never once woken up wanting to build agents. I wake up wanting to build marketing that works for the operators I serve, and the agent work is downstream of that.
The system behind the posts
What I actually build has four parts. Ad platforms. CRMs. Websites. And automations that reach past the screen into the offline world, including physical mail.
Those are not four separate tools I happen to run. They are one system, wired together, so a signal in one place moves the right lever in another. A form fills on the website. The CRM records it and decides what it is. The ad platforms hear about it and stop paying to re-acquire someone who already raised their hand. If the record goes quiet, something offline goes out in the mail. That wiring is the product. The individual tools are commodities. The way they talk to each other is not.
The crash story, the fence, the pixel: those are what it takes to make that wiring honest. They only read like AI-infrastructure posts because the hard part of the plumbing happens to be AI-shaped right now.
What account-based marketing actually means
Account-based marketing was built for enterprise. The idea is old and simple. Instead of spraying the whole market and hoping, you pick the accounts that actually matter, and you coordinate every touch around them. One company, one named target, a marketing and sales motion built to win that specific account.
I run that same discipline for a business-to-consumer sale. The twist is the size of the account. In real estate and lending the account is not a Fortune 500 logo. It is one listing. One borrower. One household. There are a lot more of them, and each one is smaller, so the coordination has to be automated to survive.
Most consumer marketing treats people as a funnel. Pour volume in the top, measure what drips out the bottom. I treat each one like an account somebody actually decided to win. That is the whole bet, and it is why the engineering matters. You cannot run account-level discipline across thousands of small accounts by hand and keep it honest.
Why the AI showed up at all
Not because AI is exciting. Because a team of a few people cannot run that many coordinated plays, across that many platforms, for that many small accounts, without something doing the repetitive judgment reliably.
That is what the agents are for. And the moment you let software take real action on a client’s ad spend and a client’s contacts, the interesting problem stops being the model and starts being trust. Idempotent writes, so a retry doesn’t double-charge a budget or double-write a contact. An audit trail, so I can say what the system did and why. A human in the loop before anything risky commits. Those are not the glamorous parts. They are the parts that keep the marketing system from lying to a client.
So the boring engineering is the whole trust layer. I write about it because it is the part that breaks, and the part that breaks is the part worth writing about.
What I’ll show you, and what I won’t
Here is the honest boundary, and I’d rather name it than pretend it isn’t there.
I’ll show you the engineering discipline. The decisions I reversed. The night the machine taught me where guardrails belong. The gate that failed correctly on a live site where the tag was present but never fired. That’s real, and sharing it is how you earn trust with someone technical.
I won’t hand you the exact plays. The specific triggers, the audiences, the sequences that make a campaign work for a particular kind of client. That is the operator’s edge, and it’s mine too. A good agency does not publish its playbook, and I’m not going to either. Naming the line is not coyness. It is the difference between a blog that helps you trust me and a blog that helps a competitor skip the eight months.
The work
The AI was never the headline. The headline is that a listing, or a borrower, gets treated like an account somebody decided to win, and the system behind that has to be honest enough to run on a real client’s money.
That is the work. The agents are just how a small team pulls it off.