I Don't Use AI. I Run It.

Most people have a chatbot. I have a staff.

By 7 AM on most mornings, I've already read a briefing. Overnight news relevant to what I'm working on, my calendar, open tasks, anything that moved while I was asleep. A scheduled process ran at 6:45, pulled the data, wrote the summary, and put it somewhere I'd see it. I didn't write a prompt for it.

I hold down a full-time hybrid job at a marketing agency while building a solo consulting practice on the side. One person, no team. What I have instead is a small fleet of agents running on hardware in my home office, each with a defined role, handling work while I'm in meetings or otherwise occupied. One coordinates, one builds, one researches and writes. Information moves between them intentionally, task by task, and every piece of work has a record I can come back to.

The memory problem

Most people who set up something like this hit the same wall eventually: the agents forget everything between sessions. Every conversation starts cold.

I solved this by giving each agent access to a running log: what I've been working on, what decisions I've made, what I'm currently tracking. There's a semantic search layer underneath it, so relevant context surfaces when it's needed without me pointing at it. Two weeks away from a project and I don't have to re-explain the whole thing. The agent reads, catches up, and picks up where we left off.

What mornings actually look like

Email triage runs at 5:45 AM on local hardware, no external API, no data leaving my machine. By the time I open my inbox, everything is already sorted: urgent, informational, can wait. I spend maybe five minutes on email most mornings instead of forty.

The briefing hits at 6:45. It's not a newsletter or a curated feed; it's a synthesized summary written specifically for my day: what's active, what's scheduled, what's flagged. It reads like something a well-organized colleague would hand you walking into the office.

I didn't build these because I enjoy tinkering. I built them because triage and status checks were eating the hours I actually needed to think.

What changed

Honestly, it mostly feels normal now. That's probably the most accurate thing I can say about it.

The difference is texture. Reactive work, the kind that used to scatter a day into thirty-minute chunks, is compressed into small windows. Most of my actual hours go toward decisions and output rather than logistics.

I also have a personal CRM to track contacts, notes from conversations, and follow-ups. Nothing off the shelf fit what I needed, so I had an agent build it. That's probably the clearest version of what this setup enables: the friction of "I wish I had a tool for this" gets low enough that you just build the thing.

The limits

None of this is autonomous. The agents don't go off and make decisions; I still set direction, review outputs, and sign off on anything that matters. The system amplifies judgment, it doesn't replace it.

It also took real time to build. The setup, the memory architecture, the routing logic, the scheduling. Somewhere around a hundred hours spread over several months.

The arithmetic is fairly straightforward, though: I'm running what would otherwise require two or three people as a solo operator. That leverage compounds.

The other model

Most people use AI as a faster way to do things manually. You go to it with a question, it answers, you move on. That's useful. It's just a different thing than what I'm describing.

The model that took longer to build, and the one that actually changed how I work, is ambient rather than transactional. It runs whether or not I'm asking it to. By the time I sit down in the morning, the triage is done, the briefing is written, and whatever moved overnight is already in my notes. I didn't do any of that.

That's a model that turns AI into a staff that works for you.