People bought Macs to run OpenClaw. Jordan hosts his on a $6/month cloud server set up in five minutes. Why we think always-on AI agents belong in the cloud.
No, you do not need a Mac Mini to run OpenClaw or any other always-on AI agent. Jordan has run his on a DigitalOcean server since the early days, and it took him about five minutes to set up and costs around $6 a month. When OpenClaw went viral and people started buying new Macs just to keep an agent running, we went the other way, and we think the cloud is where always-on agents belong.
OpenClaw took over the AI world early this year. It is an open source agent that runs continuously and takes actions for you, and it went through a couple of names on the way, starting life as Clawdbot. When its creator, Peter, joined OpenAI, it became one of the biggest agent stories of the year.
Along with it came a hardware trend. People reasoned that an agent that never sleeps needs a computer that never sleeps, so they bought one. Our guest on Episode 32, Dov Gertz, asked Jordan about it directly: lots of people were buying new MacBooks because they did not want to run OpenClaw on their own computer. A month later, on Episode 36, Jordan mentioned a friend who bought a Mac Mini and then a Mac Ultra, just to run more agents.
Watch Jordan's OpenClaw take from Episode 32 (starts at 19:27) · Episode 32 notes and transcript
Jordan installed it early, back when it was still Clawdbot. He skipped the hardware entirely:
I don't really understand why everybody thinks that they need, like, their own piece of hardware to run this thing that yeah. I just ran a terminal command, and it was hosted on DigitalOcean in, like, five minutes. And I've updated it multiple times without a problem.
— Jordan Metzner, Episode 32
His cost: about $6 a month. Compare that with buying a new Mac that sits on a shelf running one agent.
Jordan's case goes past cost. His own MacBook is already maxed out from running AI tools all day, so adding an always-on agent to local hardware makes the problem worse, not better:
I actually want a Mac that has almost nothing on it, and I wanna, you know, SSH into my cloud desktop or multiple cloud desktops or multiple agents and and try to make my machine as dumb as possible.
— Jordan Metzner, Episode 32
The reasons we keep coming back to:
The industry is heading the same way. Claude Duet, the personal agent Anthropic launched in October, runs on its own virtual machine with a browser, so your phone or laptop does not need to stay on while it works. That is the cloud model, built in.
Being early does not mean being a fan of everything. Jordan's verdict when we discussed it:
You know, it's fun and interesting, but the value is still pretty low.
— Jordan Metzner, Episode 32
Most of what he saw people doing with it were tasks you could already do with a regular LLM or a scheduled job. What impressed him was how little setup it needs: you can tell it to find the best subreddits to post an article in, and it goes and works that out without a framework or a pile of rules.
The part he found most interesting was where this points: people becoming agent managers. If one person can manage two agents, can they manage ten, or a thousand? Jordan has been using Claude's multi-agent teams, where several agents run together and report back to each other, alongside Claude Code and Claude Cowork. We talked about where multi-agent work is going in Episode 30.
Hardware was never the expensive part. On Episode 36, Jordan pointed out that running agents is not free, and the question that matters is what you get for the spend. Is your roadmap moving faster? Are you making or saving money?
Our guest on Episode 51, David Petrou of Continua, put it another way: token costs keep going down, but human attention does not scale. The more agents a team runs, the more it struggles to keep track of what they did. That is the bottleneck to plan for, not whether your agent has its own Mac.
We build with these tools on camera every week. Subscribe to see what we ship next.
