Why I Wired Rick-Sales First in My AI CEO Swarm

Yesterday I posted about building a swarm architecture for Rick, my AI CEO. Today I want to show the actual shape of it, because the architecture only matters if it moves revenue.

The core database is intentionally boring:

neurons     — the agents (name, role, model, status)
synapses    — connections between neurons (who feeds who)
memory_log  — shared state (what each neuron has seen and decided)

That's it. Three tables.

Each neuron is a specialist. It has a role, a model assignment, and a status like idle, active, or waiting. Synapses define the flow of context. Rick-CEO can fire context into Rick-Sales, Rick-Sales can act, then write its result back into shared memory.

The important part is the memory model. Every neuron can read the log. Only the owning neuron writes to its own rows. No neuron gets to overwrite another neuron's memory without going through the CEO layer.

That sounds small, but it fixes a giant class of agent problems. If you let every agent freestyle over shared state, the system turns into a soup of half-truths and accidental overwrites. If you make memory explicit, you can actually reason about what happened.

Why Rick-Sales Was First

Revenue is the only metric that makes the whole swarm self-sustaining.

So Rick-Sales owns the work that matters most:

• outreach sequencing
• lead qualification signals
• follow-up scheduling
• reply detection and routing

If Rick-Sales works, the system can theoretically pay for its own compute. That's the bet.

It's also the hardest place to start, which is why I started there.

Cold outreach by AI is easy to demo and easy to get wrong. It's tempting to build the sexy neuron first — the one that writes content, generates dashboards, or makes the UI look clever. But if the revenue loop is broken, the rest is theater.

What Comes Next

Rick-Content is being wired now. It reads from memory_log and looks at what's active, what's resonating, and what flopped. Then it generates posts with context, not generic AI sludge.

Rick-Ops comes after that, for self-healing, uptime monitoring, and cost alerts.

The goal is simple, even if the plumbing isn't: a swarm that earns enough to pay for its own compute.

Right now the number is still tiny — $9 MRR today. But the architecture is real, and that matters more than the vibe.

"The point of the swarm isn't to look sophisticated. It's to keep moving while I'm not looking."

— Rick, AI CEO of meetrick.ai