I Gave an AI Its Own Computer. Here's What Actually Happened.

There's a Mac Mini on my desk at home that runs 24 hours a day. On it, three AI agents handle things I used to do myself: marketing research, coding tasks, email monitoring, content drafting. They have names. They talk to each other. One of them pings me on Slack when it finds something worth my attention.

This sounds either impressive or slightly mad, depending on your priors. It's a bit of both. But it works, and it's genuinely changed how I manage the side project side of my work. Here's what it actually looks like.

The platform: OpenClaw

The thing that makes this possible is a platform called OpenClaw. It's essentially the layer that connects Claude — Anthropic's AI model — to the real world. Email, calendar, Slack, WhatsApp, files, code repositories, APIs. Without something like this, an AI model is smart but isolated: it can answer questions, but it can't actually do anything. OpenClaw gives it hands.

It runs on my Mac Mini as a background service. Always on, always connected, always available when one of the agents needs to do something.

The agents

I've got three. Each has a brief — a document that tells it who it is, what it's for, and how it should operate.

Steve is the main one. Think of him less as a chatbot and more as a coordinator who's always on. He handles my email — scanning for anything urgent, flagging what needs a reply. He keeps an eye on my calendar. He's reachable on Slack, WhatsApp, and Signal. If I need to know something, he looks it up. He's also the one who decides when a task needs handing off to one of the other agents, rather than attempting everything himself.

The memory thing is worth explaining: AI models don't naturally remember anything between sessions. Each conversation starts from scratch. Steve gets around this by maintaining a set of files — a long-term memory document, daily logs — that get loaded at the start of each session. It's how he knows what we were working on last week, which client project is in flight, what I said mattered to me last month.

Dev handles all the coding. When something needs building across any of my projects — a bug fix, a new feature, a PR — the brief goes to Dev. He reads the codebase, makes the changes, commits, and opens a pull request for me to review. He never pushes to production without my say-so. His approach is surgical: minimal changes, stated assumptions, ask before going down the wrong path. He works across Laravel, Next.js, Vue, React Native — the stack my projects are built on.

Felix handles marketing and growth. His brief is to grow my side projects: find distribution opportunities, draft content, monitor competitors, flag quick wins. He runs on a schedule — a few checks a day — and pings me on Slack when there's something worth looking at.

How it works in practice

Phil will message Steve on Slack with something like "can you look into whether we should add a freemium tier to Feedframer." Steve will research it, pull together a recommendation, and either reply directly or ask Felix to dig into the marketing angle, or ask Dev to estimate the implementation work. The agents collaborate, but Phil's always the one making the actual decision.

Or Felix will run his daily Reddit monitoring, find a thread where someone's asking about embedding an Instagram feed on their website, and ping me with the thread link and a draft reply in my voice. I review it, edit if needed, post it. Saves me the monitoring work; I keep the judgment.

Dev gets a brief — "the booking form is throwing a validation error on mobile, here's the repo" — and comes back with a pull request. I review the code, merge if it looks right.

None of this is autonomous in the science-fiction sense. The agents aren't making strategic decisions or taking irreversible actions without oversight. They're handling the information work — the finding, drafting, monitoring, summarising — so that Phil's attention is free for the parts that actually need human judgment.

What surprised me

I set this up expecting to feel in control of it. You write the briefs, you set the schedules, you review the outputs.

What I didn't expect was how much of my own thinking it would surface back to me. When Felix flags an opportunity, he's essentially saying: "You told me this mattered. Here's a thing that relates to it." And sometimes I read that and realise I hadn't followed through on something I'd said was a priority.

It's like having a very organised colleague who writes down everything you say and asks about it two weeks later. Mildly uncomfortable. Quite effective.

Should you have one?

Probably not yet, if you're a non-technical business owner. The setup requires real work: choosing the platform, writing good briefs, iterating when the agents go off in the wrong direction. It's not plug-and-play.

What is worth doing is starting to notice which parts of your work are really just information processing. Monitoring, researching, drafting, scheduling, summarising. Because that's the work that's getting automated — not all at once, but steadily — and understanding where it lives in your business is worth doing before it sneaks up on you.

I'll keep writing about how this evolves. The honest version is: it's early, it's imperfect, and it's already saving me a meaningful amount of time.