Client DNA: How a Siloed Knowledge Base Made 64 Landing Pages a One-Ask Job

The most useful thing Bryan Fikes figured out about AI is not that it can write or design. Plenty of tools can do that. What he figured out is how to make it produce work that stays inside a single client's voice, every time, without him having to re-explain the brand on every request.

He calls the raw material for that client DNA.

"Not only is the entire ecosystem a brain, you can have individual subsections of that that are siloed for specific purposes. And in my purpose, it's a client."

That insight — that a large AI system can be partitioned into dedicated, client-specific brains — is what turns a general-purpose tool into a marketing operation that scales without losing fidelity.

Loading the DNA

The method is straightforward to describe and demanding to execute. For each client, Fikes builds a knowledge base loaded with as much of that client's identity as he can capture — their entire DNA. Once it is in there, it changes how every output behaves.

"That knowledge base of what that client is, their entire DNA — I figured out that if I can get as much of that DNA in there, every time I go to produce something, it has that DNA as a kind of a call signal, and it produces within the scope."

The phrase "call signal" is the key. The brand identity isn't something he has to remember to mention on each task. It is baked into the silo, so every piece of output emerges already on-brand. The system doesn't drift toward generic, because the generic was walled off the moment the DNA went in.

This is the answer to the most common complaint about AI-generated marketing — that it all sounds the same. It sounds the same when the model has nothing specific to draw on. Give it a fully loaded client silo, and the output comes back inside the lines.

What it makes possible: 64 landing pages

The payoff is a change in scale that would have been unthinkable under the old model.

"All of a sudden you're like, wait a minute, I can create 64 landing pages for a client, for a campaign that's going after multiple areas. I mean, that manually that would have taken weeks, if not months."

A multi-area campaign needs distinct pages for distinct locations, each one specific yet consistent with the brand. Done by hand, that is weeks or months of work. With the client's DNA already loaded into a silo and a proven workflow in place, it becomes a single ask: bring up the playbook, run the workflow, and the pages come out on-brand and usable.

"Now I can simply just ask Bodhi to bring up my playbook, my workflow."

The scale is not the point in itself. The point is that the scale arrives without sacrificing brand consistency. Sixty-four pages, all of them sounding like the client, produced in a fraction of the time.

The Ralph loop: an intelligence layer that keeps learning

A siloed brain that never improves would still be useful, but Fikes built something that compounds. He describes it through what he calls a Ralph loop — an intelligence layer that feeds results back into the system.

"Once you put something into play and get it working really well, you can then focus on other things and then come bring it back in. Well, that intelligence layer, that putting it back in, is what has made it."

The mechanism is daily and cumulative. Every keystroke he makes and every conversation he chooses goes back into the brain, and the brain keeps evolving. The effect, in his telling, is developmental.

"It went from an infant to now — well, let's call it a moderate teenager going into, to pass its college years."

That is what an intelligence layer buys: continuous learning that turns today's output into tomorrow's baseline. When he first started, his agents wouldn't all know the playbook for a periodontal-implant dentist — the age to target, the most critical first thing a potential customer needs. Now they all know it. The system's overall knowledge rises every time work flows back through the loop. Compounding output is not a metaphor here. It is the architecture.

Smoke-test before you trust it

None of this works if you simply trust whatever the system produces. Fikes is emphatic that the output has to be validated before it is relied on, and he uses a term he picked up from the vibe-coding world: smoke testing.

"Knowing that I've smoked out — or smoke-tested — all these new AI terms when you're vibe coding. But the output of what you actually get is something that's actually tangible. You can actually use it."

The discipline is to test a workflow before you trust its output at scale. You don't fire off 64 landing pages on faith. You prove the workflow produces tangible, usable results first, then set it loose. This is the safeguard that keeps the speed honest. Smoke testing is what stands between an impressive demo and an embarrassing mistake delivered to a real client.

The whole team gets smarter together

The Ralph loop does something beyond improving a single silo — it raises the knowledge of the entire agent system. Fikes illustrates it with the example he returns to often: the playbook for a periodontal-implant dentist, with all its nuances about which age group to target and what a potential patient needs first.

"When I first started out, if I said, what's the playbook for that periodontal-implant dentist, they wouldn't all know. The other agents wouldn't know the answer to that. Now they all know the answer."

That is the compounding effect made concrete. Knowledge captured in the course of serving one client doesn't stay stranded with one agent. It flows through the loop and lifts the whole system's competence. The more work runs through, the more the entire team knows.

And the measure of whether it is working is simple and unsentimental. Fikes ties the word "success" directly to output:

"That's why it's becoming the word success — because it's successfully producing the output that I want it to produce."

A system that learns is only valuable if it learns toward the result you actually need. The Ralph loop, fed by smoke-tested workflows and loaded client DNA, does exactly that.

Why this is the real moat

It is tempting to think the advantage here is the AI itself. It isn't — anyone can access the same models. The advantage is the method: a fully loaded client DNA silo, a proven and smoke-tested workflow, and a Ralph loop that makes the whole system smarter every day.

That combination is what lets one operator deliver agency-scale output that never loses the client's voice. The model is the raw capability. The client DNA gives it focus, the workflow gives it reliability, and the intelligence layer gives it memory. Put together, they answer the question every business owner asks about AI marketing: how do I get the scale without the slop?

You silo the DNA. You smoke-test the workflow. And you let the loop keep learning.

---

Want every piece of your marketing to come back sounding exactly like your brand — at a scale you couldn't reach by hand? Bryan Fikes builds siloed, DNA-loaded workflows for a small number of clients who care about both speed and fidelity. Schedule a strategy session with Bryan to see how it would work for you.