Six Months Running a Wolf Pack: What I Learned Building a Business on AI Agents

Six months ago I made a decision that felt insane at the time. I was going to stop trying to hire my way out of operational problems and start building agents to handle them instead.

I did not call them agents at first. I called them bots, assistants, whatever made sense in the moment. But over time, nine of them took shape with names, roles, and clear ownership of tasks. I started calling them the Wolf Pack. Atlas writes. Loki sends outreach. Echo handles replies. Roki does research. Shakti publishes. Vega distributes on LinkedIn. Drake pushes to Devon’s publisher. Briefing reports to me every morning. Osito handles the one thing I let myself play: Last War.

This is what six months of building that system actually looked like.

What I Got Right From the Start

The best decision I made early was treating each agent like a person with a job, not a tool with a prompt. That meant giving each one a SOUL.md file. Identity. Voice. Hard rules. A clear scope of what they own and what they never touch.

When Atlas writes LinkedIn posts, it does not guess about brand voice. It reads its identity file first. When Loki drafts outreach, it knows exactly which firms to target and which language is off-limits. That specificity is what separates an agent that helps from an agent that creates cleanup work.

The second thing I got right: I gave each agent a lane and made lane violations expensive. An agent that starts doing things outside its scope becomes unpredictable. Unpredictable agents cost time. The whole point is to save it.

What Failed in Months One and Two

I underestimated context drift. An agent that works perfectly in one session starts making odd decisions three sessions later if you have not given it a way to stay oriented. I had to build a memory system. Daily notes. Lazy-load context files organized by topic. A MEMORY.md that agents could search on demand instead of loading everything up front.

That was not fun to build. But once it worked, everything downstream got more consistent.

I also tried to run too many agents in parallel too fast. There is a ceiling on how much parallel work produces coherent outputs. Push past it and you get agents that contradict each other, step on each other’s work, or produce content that feels like it came from different companies. You have to build one agent fully before you start the next one.

The Intake Lesson I Kept Coming Back To

I built eNZeTi because I saw a specific problem up close: law firms were losing cases at intake. Not because their coordinators were bad. Because no one was in their corner during the calls that mattered. No real-time support. No coaching. Just a person on the phone doing their best with whatever training they happened to get six months ago.

Building the Wolf Pack taught me the same lesson from the other side. My agents do not fail because they are built wrong. They drift when I do not give them the right support structure. Identity files. Memory. Clear scope. Feedback loops. The agent is never the problem. The support gap is the problem.

I think about that every time someone asks me whether AI should replace human teams. The answer is the same in both directions. Replace nothing. Equip everything. The humans in your law firm are not the reason intake breaks down. The support gap is. That is the only thing worth fixing.

Month Three: When the System Started Working

The shift happened when I stopped building agents and started managing them. There is a real difference.

Building means you are in the system constantly, adjusting prompts, testing outputs, cleaning up errors. Managing means the system runs without you until something breaks. You review outputs. You course correct. You update identity files when the business evolves. But you are no longer in the loop on every task.

Month three was the first month where I woke up to a full briefing from my reporting agent, saw that outreach had gone out, LinkedIn posts had published, and articles had been saved for review, and I had not touched any of it the night before. That was the first time it felt like a real team.

What the Numbers Look Like

I am not going to fabricate a tidy before-and-after table. What I can tell you is this: I am producing more content, more outreach, and more distribution across more channels than I was six months ago, with less time in execution. The time I spent on execution is now spent on direction. Writing better briefs. Reviewing outputs. Adjusting strategy based on what is actually working.

That shift, from execution to direction, is worth more than any productivity percentage I could name.

What I Would Tell a Founder Starting Today

Start with one agent. Give it a complete identity. Define its lane with enough precision that you could explain it in one sentence. Run it until it is boring. Then build the next one.

Do not skip the memory architecture. It feels like overhead until the day you realize your agents are inconsistent and you cannot figure out why. Memory is what makes consistency possible at scale.

And do not confuse automation with augmentation. The difference matters more than most founders realize. Automation removes the human from the process. Augmentation makes the human in the process sharper. The Wolf Pack does not run my business without me. It makes my judgment better by handling everything that does not require my judgment.

That is the model I brought into eNZeTi. The intake coordinator does not get replaced. They get backed by a system that puts the right response on their screen the moment a prospect hesitates. Same human. Same empathy. Better support.

Six months in, I would not go back to the way it was before. Not because the agents are perfect. Because the gap between me and the work I actually need to do has never been smaller.