The AI company for Atlanta businesses
ServingMetro Atlanta
StatusBooking calls

Articles · Watch us build · 4 min read

How a 24/7 AI workflow actually works

What an always-on AI agent is, the easy setup for non-technical owners, and the live test where it set up its own email service and sent me a morning brief

In January I spent three weeks with a tool that changed how I thought about AI. Back then everybody called it Clawdbot. Today it's OpenClaw. I watched a video by Alex Finn, and before I knew it I was setting it up.

Now, I'm not that coding dude. I'd been doing this for about three weeks. So this is the non-technical version of a 24/7 AI workflow: what it is, what you need, the easiest setup, and what happened when I tested it live.

What it actually is

It's an AI system that takes over a computer, and you talk to it like a regular person. It builds apps, sends emails, writes tweets, writes code.

When I recorded this, the features that mattered were:

  • You talk to it through WhatsApp, iMessage or Telegram
  • It has persistent memory, which Claude, Gemini and ChatGPT were all struggling with
  • It can take over a browser
  • It has full system access, plus skills and plugins

Setup was copying one command into your terminal and walking through an easy onboarding.

Two pieces of advice before you do that.

Use your subscription, not the API. If you're on something like Claude Max or a Google Ultra plan, point it at that. API charges add up.

Do not put it on your personal computer. When I say this thing is powerful, it is powerful. Full system access means it can get into everything. Run it on a separate computer, an old one lying around if you have it. You don't want to hand it your own stuff, because it can do some scary stuff.

The easiest workflow for non-technical people

I named mine Brock and talked to it over Telegram. It talks back like a person. I'd tell it what we were about to do and it would say, "Sounds good. Standing by."

Here's the workflow I'd give anybody who isn't technical. The "proper" way to let it work in your accounts is API connections, authorization, tokens. That stuff breaks. I've run into those problems. AI isn't perfect yet.

The easy way: let it open its own browser. Then go into that browser, log in to the accounts you want it to use, and pin those tabs. For me that was GitHub, my email, my websites, NotebookLM, and the Gemini chats where I draft scripts for videos. Already logged in, so it isn't asking for keys and authorization every time.

That's how I had it set up my whole X account. It made the account and started sending out tweets. I still had to get it to work on the banner. But regular people like me don't know how to wire up an API connection to X. I opened a tab.

Same rule as the computer, though. A pinned, logged-in tab means it can act as you in that account. Only pin what you'd hand to a new employee.

The live test: a morning brief, built from scratch

The morning brief idea came from Alex Finn's video.

I messaged Brock on Telegram: can you send me the morning brief we talked about? It came back with the weather (too cold) and a rundown of what it did overnight. Yes, overnight. It had worked on an app called Budget Quest, launched a Patreon, and lined up the next thing we'd talked about, a gamified daily checklist app, starting with the product requirements doc.

Then I gave it something harder. I had the Resend page open in its browser. I told it: set this up from scratch, get the key, and send the morning brief to my email.

It went to work. It created the API key itself. I won't show you that part. Then I checked my inbox.

"Good morning, Richard." Sunday. Atlanta, twenty-five degrees. Today's focus. "Have a productive day, Brock."

It set up the account, got the key it needed, and sent the email. All I did was open a tab and send one message on Telegram.

What I'd tell you about it

It's a 24/7 employee in the sense that it keeps working when you stop. It's still an employee, though. It needs the right access, its own computer, and somebody checking the work. It isn't ready to take over as a human.

And here's what I said then that I still believe. Tools like this will take the jobs of people who don't want to learn anything. The people who learn to work with them are the ones who come out ahead. In a few weeks I went from just being a dude to building software. Any one of you can do it. You just have to learn.

What this means for you

If you want an agent working around the clock inside your business, the first question is where your files and your data go. That's why RMDW built Jah 6.0, our own AI model: its weights are public on Hugging Face, it powers Kiyomi, our private AI for offices with sensitive files, and it plugs into tools like Claude Code, Codex and Cursor.

More in Watch us build