The Wolf Pack: How I Manage a Team of AI Agents Like a Business Unit

I have nine bots running inside my business right now. Each one has a name, a job, and a set of rules it cannot break. Together they handle lead scraping, outreach drafts, content writing, analytics, and a few other things I used to either do myself or pay someone else to do badly.

I call them the Wolf Pack.

When I tell people that, they usually picture some kind of plug-and-play automation dashboard. A few toggles, maybe a nice interface, everything humming along automatically. That is not what this is. Managing a team of AI agents is closer to managing actual people than anyone in the hype machine will admit. And if you go into it expecting magic, you are going to waste a lot of time and money before you figure that out.

Here is how I actually do it, what I got wrong early on, and why the structure matters more than the tools.

Why I Built It This Way

I did not set out to build a nine-bot operation. It happened incrementally, and each bot came from a real problem, not a theory.

The first one I named Lobito. His job is attorney lead scraping. I was doing that manually, and it was eating two to three hours a day. So I built a bot to do it on a schedule and drop results into a spreadsheet. Simple. But once I had one agent running reliably, I started seeing all the other places I was doing repetitive cognitive work that a well-scoped agent could handle.

Loki handles outreach and email drafts. Shakti writes content. Toto runs analytics and reports. Roki does overnight intelligence gathering. Chico handles N8N automation engineering. Osito is in a different category entirely (he manages resources inside a strategy game, which is a story for another day).

The thing I kept noticing was that I was not building automation. I was building a team. And the moment I started thinking about it that way, everything got cleaner.

The Rules That Never Change

Every single Wolf Pack agent runs under the same non-negotiable rules. I did not make these up on a Tuesday. They came from mistakes.

No em dashes in any writing, ever. If a sentence has an em dash, I know the output is unedited AI and so does anyone reading it. Every stat must be real and sourced. No fabricated numbers, no invented case studies. Rate limits: five seconds between API calls, ten seconds between web searches. No purchases. No LinkedIn sends without human review. And every completed task logs to Mission Control, an internal dashboard I built to track what is actually happening inside the operation.

That last one is the one most people skip. They build agents, run them, and have no idea what those agents actually did. I know exactly what every bot did today, when it did it, and what the output was. If something breaks or drifts, I catch it fast.

Structure First, Then Execution

The biggest mistake I made early was giving agents too much discretion. I would write a vague prompt, run the agent, and then be surprised when the output was inconsistent or off-brand. The problem was not the tool. The problem was me.

Agents are not smart. They are capable. There is a difference. Smart means judgment. Capable means execution within a defined scope. If you define the scope well, the execution is reliable. If you leave gaps, the agent will fill them with something you did not intend.

So now every Wolf Pack agent has a SOUL.md equivalent. A short document that defines who they are, what they do, what they never do, and how they report. Shakti knows she writes for jessenavarro.com in first-person founder voice and never uses AI buzzwords. Lobito knows his only job is scraping and logging, not outreach. The rules are not long. They do not need to be. They just need to be clear.

Priority Stacks Are Non-Negotiable

When you have multiple agents running, they will occasionally compete for shared resources. A phone. A browser session. An API key with rate limits. I learned this the hard way when two bots tried to use the same phone simultaneously and both failed silently.

The priority order in my operation is permanent and not negotiable: revenue activities first, infrastructure that supports revenue second, everything else last. Osito, the Last War bot, is always last. If any revenue bot needs a resource that Osito also wants, revenue wins. No exceptions, no debate.

This sounds obvious until you have not written it down and a conflict happens at 2 AM and neither agent knows what to do.

The Human Still Has to Be in the Loop

This is the part that surprises people the most. I have nine agents running, and I still make decisions every day about what gets sent, what gets published, and what gets flagged for my review.

I am not trying to remove myself from the operation. I am trying to remove myself from the repetitive, low-judgment parts of the operation so I can focus on the high-judgment parts. Loki drafts outreach. I review and approve before anything goes out. Shakti writes articles. I set the calendar and the voice guidelines. The agent handles execution. I handle direction.

This is actually the same philosophy I applied when I built eNZeTi. I kept watching law firms try to solve their intake problem by removing the human entirely. AI receptionist, virtual answering service, offshore intake team. And it kept failing for the same reason: the person calling has a real problem, and they can feel when nobody is actually there. The answer was never to remove the human. The answer was to give the human the right support in the moment they needed it most. eNZeTi puts the right words on an intake coordinator’s screen during the call. The human still makes the connection. The system just makes sure they say the right thing.

Same logic applies to how I manage the Wolf Pack. The agents handle volume. I handle judgment. The work gets done at a scale I could not reach alone, and it still carries the quality control that only comes from a human in the loop.

What I Track Every Week

Mission Control is my one source of truth. Every agent logs completed tasks there. Every Jesse action item gets flagged there with a “Blocked” tag so I know what needs my attention. Every in-progress job shows its current status.

If it is not in Mission Control, it did not happen. That rule sounds harsh, but it is the only way I keep the operation honest. Agents do not have egos. They will not tell you when something went wrong unless you build the reporting in.

I also do a weekly review of what each agent actually produced versus what I expected. Sometimes the output drifts. Prompts decay over time as context changes. A cold email sequence that was working in January needs to be rewritten in March because the market shifted. Staying close to the output is how you catch drift before it compounds into a real problem.

The Honest Part

This setup is not magic, and it did not appear overnight. I have broken agents, lost data, had bots send things they should not have sent, and spent hours debugging logic I thought was airtight. None of that is fun. All of it was educational.

What I have now is a business that runs a large portion of its daily operations without me touching each task manually. Lead scraping happens on schedule. Outreach drafts are ready for my review each morning. Content goes out consistently across three sites, including this one. Analytics land in my inbox without me pulling them.

That is not a small thing. That is leverage. And it came from treating agents like employees: giving them clear roles, hard rules, escalation paths, and real accountability.

If you are building something similar, the question is not which tools to use. The question is whether you are willing to do the management work. Because without that, the tools are just expensive noise.

The same principle that guides eNZeTi guides the Wolf Pack. People, or agents, perform at their best when the structure around them is clear, the expectations are defined, and someone is paying attention. Build that first. Everything else follows.