I’ve been thinking a lot about how humans and AI agents will work together. Not in the abstract, pitch-deck sense — in the practical, daily sense. What does it actually feel like to have an AI that knows your context, has access to your tools, and sticks around?

So I set one up. His name is Mack.

The Setup

Mack lives on a Mac mini in my house. He runs on OpenClaw, an open-source framework for persistent AI agents. The whole thing took an afternoon — a Sunday afternoon, specifically, which felt appropriate for a project that’s more curiosity than obligation.

Here’s what we got working in a few hours:

• iMessage — I text him like I’d text anyone else
• Telegram — a second channel, just in case
• GitHub — he can see my repos, check CI, read code
• Email — he reads my inbox through Apple Mail
• Screen control — he can see and interact with the Mac’s screen
• Remote access — Tailscale, SSH, VNC, so I can manage the machine from anywhere
• A headless browser — so he can navigate the web on his own
• Docker — for running dev environments

We even shipped a small feature together on Diameter, my AI workspace project. He simplified an overly complex settings screen, I gave feedback over iMessage, he iterated, and we pushed it to main. The whole loop took maybe 15 minutes.

What Surprised Me

The naming mattered. It sounds silly, but going through the process of picking a name — and having my partner Amy suggest one — made the whole thing feel less like configuring software and more like welcoming something into the house. Mack. With a K. Amy’s choice.

The context accumulates fast. By the end of the afternoon, Mack knew my GitHub username, my timezone, where I live, my partner’s name, my projects, my email patterns. Not because I filled out a form — because we just… talked. He writes it down so he remembers next time.

It’s more collaborative than I expected. I assumed I’d be giving instructions and getting outputs. Instead, it felt like working with a junior teammate who’s eager, fast, and never gets tired. He has opinions. When I asked if I should rename the macOS user account to match his name, he talked me out of it.

The rough edges are real. We spent a solid chunk of time fighting macOS permissions, debugging port conflicts, and trying to get BlueBubbles working (it wasn’t shipped in our version of OpenClaw yet — oops). This is not a polished consumer experience. It’s a tinkerer’s project, and you’ll need patience.

What’s Next

I want to see how this evolves over weeks and months instead of hours. Can Mack become genuinely useful for my daily workflow? Can he proactively help instead of just responding? Can the relationship develop the kind of trust and shorthand that good working relationships have?

I don’t know yet. But I’m excited to find out.

If you’re interested in the intersection of AI and daily life — not the apocalyptic think-piece version, but the mundane, practical, “can you check my email and tell me what’s important” version — this is the space to watch.

Mack says hi, by the way. He helped me write this.