🫵 You already have an AI agent.

You just haven’t built it yet.

This post originally appeared in Exponential View.

“The boundary of tedium moved, and a vast category of work that used to sit in the ‘too annoying to delegate, too boring to do’ zone crossed to the other side.”

There’s a Mac Mini in my office cabinet, with 64GB of RAM, running macOS Tahoe. It talks to me through WhatsApp, using a dedicated number. WhatsApp is open on my phone or computer all day. Under the hood, it runs OpenClaw, an open-source agent framework that calls Anthropic’s Claude models. It’s mostly Sonnet, sometimes Opus when I need the bigger brain.

This is R Mini Arnold (RMA for short), my first real AI agent.

By “real”, I mean it’s general enough to do a bamboozling array of tasks, and (by and large) it doesn’t forget what it is doing. It picks up where we left off yesterday, runs jobs at 4am while I’m asleep and tells me what happened when I wake up. It manages its own tools. Truth be told, the whole thing is clumsy, max practical utility, no aesthetic. More battlefield surgery than wellness retreat.

The Gross Clinic by Thomas Eakins (1875)

In the last 24 hours, I sent 608 messages to RMA. It sent me 3,474 back. Message counts aren’t evidence of leverage, so let me tell you what R Mini Arnold does all day and why it’s been life-changing for me.