DiveInA 2-person team doesn't have time to run marketing. So we built one that runs itself.
Okeanos installs pools in Ontario and wants to 4x its volume. Two people handle marketing, sales, follow-up, ads, and content, on top of running real installs. The bottleneck was never visibility. It was time.
We designed DiveIn, a system of AI agents that runs the marketing function and reports to the two people who own it. A CMO agent sets the plan, reviews the work, and checks in over Telegram.
Watch DiveIn work
Problem
Two people, the workload of a company
Okeanos wants to 4x its installs. But two people run marketing, sales, ads, content, and follow-up, while overseeing real installs. There's a ceiling, and it's time.
Leads go cold, opportunities lost
A lead comes in and goes cold before anyone replies. Ad budget keeps spending on what isn't working. Content never gets made. Not because no one cares. Because there's no one left.
Solution
Meet DiveIn, a 15-agent agentic marketing department
DiveIn is a 15-agent AI marketing team that runs the function for a two-person company: scoring leads, sending follow-ups, shifting ad budget, and reporting through Telegram like a coworker.
How it's structured
15 agents, each with one job
One scores leads, one writes social, one watches ad spend. Above them sits a CMO agent that acts like a real marketing lead. It sets the weekly plan, reviews every agent's output, catches conflicts, and brings the humans in when it matters.
Almost everything automated, human in the loop
A lead comes in, gets scored and routed. A follow-up goes out in minutes. A campaign tanks, the budget moves. All of it without a human, because two people don't have time to approve every email. That was the point.
The line is risk. A bad Google review comes in, an agent drafts the reply. A human reads it before it goes live. You don't let AI handle someone's frustration unsupervised.
What I learned
Build a dial, not a switch
We chose full autonomy for them: 15 agents, go. Teams that placed higher gave the business a toggle, automatic or manual, and let them choose what to hand over. We decided for the client instead of letting them decide.
Match the build to what they're ready for
Okeanos was already running 6 agents with guardrails. We jumped them to 15 autonomous ones. That's not the next step, it's three. The system wasn't too weak. The org wasn't ready for that much at once.
Pitch the experience, not the architecture
We led with the stack because it seemed impressive: 15 agents, Telegram, coordination layers. They weren't evaluating a tech stack. They were asking whether it helps their company. The most impressive system and the right one aren't the same thing.
What I'd do differently next time
The agents work. I'm proud of the system. But if I ran this again, I wouldn't open Cursor first. I'd map the customer journey, be honest about what the company is ready for, and decide the right mix of human and machine before writing a line of code. Then build agents. Probably fewer.