← Back to Blog
The Power-Cut Test: How Founder-Led Service Businesses Build AI Workflows That Keep Delivering When the Wi-Fi Dies, the Laptop Closes, and You're Asleep

The Power-Cut Test: How Founder-Led Service Businesses Build AI Workflows That Keep Delivering When the Wi-Fi Dies, the Laptop Closes, and You're Asleep

November 22, 2026·7 min read

I was on a call in Canggu when the power went out. Not a flicker. The fan, the router, the monitor, all gone at once, and the street went quiet in that specific way it does when forty other routers die at the same moment.

I finished the call on my phone. Then I sat there and asked myself the question I now ask every founder I work with: if this lasted six hours, how much client work would still have moved?

What the power-cut test actually measures

Most people hear that story and think about batteries. Buy a UPS, order a Starlink dish, keep a local SIM in a drawer. All useful, all cheap, all solvable in an afternoon.

The power cut is not a hardware test. It is a test of where your business physically lives. For most founder-led agencies I know, the answer is uncomfortable: inside the founder, on one laptop, in a folder called Drafts, held together by the fact that somebody is awake and reachable.

I have been building businesses since 1995 and teaching since 1998, and I have trained more than 90,000 people. The pattern is the same whether the founder is in Bali, Chiang Mai, or Lisbon. The internet is rarely the real constraint. The real constraint is that every deliverable passes through a checkpoint that is literally one person, and that person is on a scooter.

Your laptop is the server, and that is the actual problem

Here is the reframe that changed how I build: your laptop is the server. Not as a metaphor, as the literal architecture. The brief lives in your Downloads folder. The client's brand voice lives in your head. The draft goes out when you remember to send it. Every one of those is a single point of failure with a battery inside it.

Cloud-based AI agent workflows fix this by moving the whole chain to places you never have to carry. Concretely, that is four moves.

  1. Pick one source of truth. A shared Google Drive folder or a Notion database per client, holding the brief, the brand docs, past deliverables, and call transcripts. If a document lives on your desktop, it does not exist as far as an agent is concerned.
  2. Put a queue in front of the work. An Airtable base with one row per deliverable, a status field, a due date, and a link to the brief. The queue becomes the thing everyone looks at instead of you.
  3. Wire the trigger. n8n or Make.com, both cloud-hosted, watches the status field. When a row moves to Ready, the workflow pulls the brief, the brand doc, and two past examples, sends them to Claude or GPT, and writes the draft back onto the row.
  4. Notify in Slack, not email. One channel per client, a short message when a draft lands. Email buries things. Slack makes them feel like they happened.

Once that is running, closing the laptop is an inconvenience instead of an outage. The work sits in a queue in someone else's data center, which is exactly where you want it while you are on a boat to Nusa Penida.

Async is not a scheduling preference, it is the substrate

What's becoming clearer to me every time I build one of these is that async client operations come first, because an agent can only work on text. A decision you made on a call and never wrote down is invisible to every tool you own.

That is why so many founders buy the automation software, connect two apps, and get nothing back. There is no written record for the machine to act on.

So build the async layer before the automation layer.

  • Record every client call and auto-file the transcript. Fathom or Fireflies joined to the calendar, with n8n dropping the summary into the client's folder the moment the call ends.
  • Write decisions down where the client can see them. One shared doc per client, dated bullets. It takes me a minute and it kills the whole "wait, what did we agree" email chain.
  • Replace status meetings with one written update, same day, same format, every week. Clients relax when they can read the state of things. Relaxed clients stop needing you on the phone.
  • Use Loom for anything that genuinely needs a face and a screen. Two minutes of video is still async, and it keeps the warmth that pure text loses.

After a month you have a corpus: briefs, transcripts, decisions, past work. That corpus is what the agents feed on. The async system and the automation system are not two projects. They are one project, and this is the order they go in.

The approval step is the only place you should still be standing

Automating client deliverables without your laptop does not mean shipping unreviewed work. It means the review is a designed step instead of an accident of your calendar.

I like one checkpoint and only one. The agent produces the draft, the draft lands on the queue, and a named human presses approve. Everything before that press is automated, and everything after it is automated too: the file gets formatted, the email gets written, the client gets notified, the row gets closed. There is a longer walkthrough of how I wire this at https://workshop.mastermindshq.business.

Two rules make that survive a dead router.

The approver cannot be you by default. This is the part founders resist hardest, and it is the whole point. If the only person who can approve the work is the founder, you have built a faster version of the bottleneck, not a way out of it. Pick the person who has been doing the second pass for six months and give them the button.

The approval has to be one tap from a phone. If approving means opening a laptop, finding the file, and rewriting three paragraphs, it will not happen at 2am in a guesthouse in Chiang Mai. If it means tapping a Slack button that fires a webhook back into Airtable, it will happen.

Build the boring redundancy, then stop thinking about it

Physical redundancy first, because it takes an afternoon. Two ways online: a local SIM with a real data plan plus a spare phone that can serve as a hotspot, and either Starlink or a coworking space with a generator. Two ways to power the laptop: a small UPS and a fully charged battery bank. None of this is interesting. All of it costs less than one lost retainer.

Then the operational version, which almost nobody does:

  • Two people who can approve work, named in writing, with access already tested.
  • Two places the files live, with one declared as source of truth and the other as sync backup.
  • One written log of what shipped and what is pending, visible to the client and not just to you.

The redundancy only counts if it has been used before it matters. My partner and I trade the approve button for a week every quarter, partly to test it and partly because the people I love and admire are the reason any of this is worth building.

The test I actually run now is boring. I close the laptop at 9pm in some city where the power is unreliable, and in the morning I read the log. If work moved while I slept, the system works. If it queued up politely and waited for me, it doesn't, and I know exactly which row to fix.

What the test is really for

The goal was never to work from a beach. It was to build something that does not stop when the power does. That distinction matters for AI operations for founder-led agencies, and it is the difference between a clever demo and a business that holds its value while you sleep, travel, or finally take a real day off.

If you want to see the two paths through this, the comparison page is at https://www.mastermindshq.business/mastermindvsl. The short version is that one path is learning it alone in the dark, and the other is building it with people who have already broken it in ten different ways.

If this resonates and you want to build this kind of infrastructure in your own business, the Mastermind is where we do the work live. You can learn more at mastermindshq.business.

Ready to put this into practice?

Join Joe Che's AI Business Mastermind, a small cohort for founder-led service businesses that want to systematize their operations with AI.